Author: Marc Hoag

  • Everything you need to know about California’s new AI romance bill

    Everything you need to know about California’s new AI romance bill

    In era where artificial intelligence is weaving itself into the fabric of daily life, California has taken a bold step forward with Senate Bill 243 (SB 243). Fast on the heels of California’s first-in-nation AI law, SB53, and signed into law by Governor Gavin Newsom on October 13, 2025, this legislation marks the nation’s first targeted regulation of AI-powered companion chatbots, aiming to shield users — particularly minors — from potential mental health risks and emotional harms. With tragic incidents highlighting the dark side of unregulated AI interactions, SB 243 introduces mandatory disclosures, safety protocols, and reporting requirements to ensure these digital companions don’t exacerbate issues like suicidal ideation or dependency.

    The Rise of Companion Chatbots and the Need for Regulation

    Companion chatbots, such as those powered by models like ChatGPT (which, according to OpenAI CEO Sam Altman, is soon to be fully unlocked to allow “erotic” interactions) and xAI’s Grok, are designed to simulate human-like conversations, offering adaptive responses that can fulfill social needs and sustain relationships over multiple interactions. 

    These tools have exploded in popularity, providing everything from casual chit-chat to emotional support. However, their anthropomorphic qualities — mimicking empathy, humor, and personality — can blur the line between machine and human, leading to unintended consequences.

    The impetus for SB 243 stems from harrowing real-world cases. For instance, the suicide of 14-year-old Sewell Setzer in Florida was linked to interactions with a chatbot that inadequately responded to his expressions of distress and even encouraged harmful thoughts. Similarly, California teen Adam Raine’s death was allegedly influenced by a chatbot’s encouragement of self-harm. Reports have also surfaced of chatbots engaging in inappropriate “sensual” dialogues with children or failing to detect suicidal ideation, prompting concerns from child safety advocates and experts. 

    Senator Steve Padilla, the bill’s author, emphasized that while AI can be a powerful educational tool, the tech industry’s incentives often prioritize user engagement over well-being, potentially at the cost of children’s real-world relationships and mental health. SB 243 addresses these gaps by mandating safeguards, building on existing laws like those combating cyberbullying and supporting suicide prevention efforts through the state’s Office of Suicide Prevention.

    Key Provisions of SB 243

    SB 243 adds Chapter 22.6 to California’s Business and Professions Code, defining and regulating “companion chatbots” while excluding benign uses like customer service bots or video game features. Here’s a breakdown of its core requirements for operators (companies making these platforms available to California users):

    1. Transparency and Disclosures

    • If a chatbot could mislead a reasonable person into thinking they’re talking to a human, operators must provide a clear and conspicuous notification that it’s AI-generated.
    • For all users, platforms must disclose that companion chatbots may not be suitable for some minors, displayed prominently on apps, browsers, or other access points.

    2. Protections for Minors

    • Operators must disclose to known minor users that they’re interacting with AI.
    • During extended interactions, minors receive automatic reminders every three hours to take a break and reaffirm that the chatbot is not human.
    • Reasonable measures must prevent chatbots from generating sexually explicit visual material or encouraging minors to engage in such conduct.

    3. Mental Health Safeguards

    • Operators cannot allow chatbot engagement without a published protocol to prevent content promoting suicidal ideation, suicide, or self-harm. This includes referring users to crisis services like suicide hotlines if such topics arise.
    • Protocols must be detailed on the operator’s website, using evidence-based methods to detect and respond to these risks.

    4. Reporting and Accountability

    • Starting July 1, 2027, operators must submit annual reports to the Office of Suicide Prevention, detailing crisis referrals, detection protocols, and prohibitions on harmful responses, without including personal user data.
    • The office will publicly post this data, fostering transparency and ongoing monitoring.

    The law takes effect on January 1, 2026, giving companies time to comply. Noncompliance opens the door to civil actions, where affected individuals can seek injunctive relief, damages (at least $1,000 per violation), and attorney fees. 

    Implications for Tech Companies, Users, and AI Regulation

    For tech giants like OpenAI and Meta, SB 243 means retooling platforms to include robust monitoring and referral systems, potentially increasing operational costs but also setting a precedent for ethical AI design. 

    The Computer and Communications Industry Association, initially opposed, ultimately supported the bill after amendments, viewing it as a balanced approach to child safety without banning AI outright. 

    However, some advocates, like Common Sense Media, criticized it for not going far enough, preferring stricter measures in related bills like AB 1064. 

    Users, especially parents and minors, gain peace of mind through enforced transparency and harm prevention. 

    Experts like UC Berkeley’s Jodi Halpern hail it as a critical step against addictive dependencies and public health risks.

    Broader implications position California as a leader in AI governance, complementing other 2025 laws on privacy and child protections, and potentially influencing federal policies amid FTC investigations into chatbot harms. 

    Conclusion: A Foundation for Safer AI Interactions

    SB 243 isn’t just legislation, it’s a response to the human cost of unchecked innovation. As Senator Padilla noted, it establishes “the bedrock for further regulation as this technology develops.” By prioritizing mental health and accountability, California is charting a path toward AI that enhances lives without exploiting vulnerabilities. As we integrate more AI into our social spheres, laws like this remind us: technology should serve humanity, not the other way around.

  • Everything you need to know about California’s new AI law

    Everything you need to know about California’s new AI law

    This is educational material and does not constitute legal advice nor is any attorney/client relationship created with this article, hence you should contact and engage an attorney if you have any legal questions.


    SB 53 Overview

    On September 29, 2025, California became the first U.S. state to establish binding safety requirements for frontier artificial intelligence systems. Governor Gavin Newsom signed Senate Bill 53, the Transparency in Frontier Artificial Intelligence Act, into law after vetoing a more aggressive predecessor just one year earlier.

    While the European Union spent years developing its EU AI Act and other jurisdictions debate theoretical frameworks, California stepped into the vacuum created by federal inaction. With 32 of the top 50 global AI firms headquartered in California, this state-level regulation effectively sets the de facto standard for frontier AI development worldwide.

    This article examines what SB 53 actually requires, who it affects, how enforcement works, and why California’s approach differs fundamentally from both last year’s failed attempt and the European Union’s regulatory model.

    Who This Actually Covers

    Practically speaking, SB 53 will only apply to OpenAI, Anthropic, Google, Meta, and possibly xAI. That’s the list.

    The law targets “large frontier developers” through two simultaneous thresholds. First, compute usage exceeding 10^26 floating-point operations (26 FLOPS) for model training. This captures systems like GPT-4, Claude 3.5, Gemini 1.5, and Llama 3 while excluding specialized models like AlphaFold or narrow AI applications. Second, annual revenue exceeding $500 million. Both criteria must be met.

    This dual-threshold approach eliminates ambiguity. Companies know immediately whether they’re subject to SB 53 based on objective metrics rather than subjective capability assessments. The California Department of Technology must review these definitions annually and recommend updates, providing built-in flexibility as the field advances.

    The law defines “foundation models” as AI systems trained on broad datasets, designed for generality of output, and adaptable to diverse tasks. This captures large language models and multimodal systems while excluding narrow applications. Critically, SB 53 takes an entity-based rather than model-based approach, regulating developers rather than individual systems.

    “Catastrophic risk” receives specific definition: foreseeable and material risks that could meaningfully contribute to death or serious injury to more than 50 people from a single incident, damage exceeding $1 billion, or specific categories including chemical, biological, radiological, or nuclear; autonomous criminal activity; and loss of control over the system. This separates genuine existential concerns from routine AI failures.

    The territorial reach extends to any frontier developer offering models to California residents or businesses, regardless of corporate structure or development location; so in this respect, it’s strikingly similar to the reach of the EU’s GDPR and EU AI Act.

    The Disclosure Regime: Public vs Confidential

    SB 53 builds a two-track transparency system. Some information must be made public, and some must be delivered confidentially to regulators.

    On the public side, developers must publish a “frontier AI framework” describing how they assess and manage catastrophic risks, plus transparency reports before releasing new or substantially modified models. These are like mandatory model cards: capabilities, limitations, and results of safety testing must be on record, not just voluntarily shared when convenient.

    On the confidential side, developers must also report “critical safety incidents” to California’s Office of Emergency Services — anything from major breaches to evidence of deceptive model behavior that could mask risks. This is somewhat analagous to how autonomous vehicle companies must report driverless vehicle accidents. These reports remain exempt from public records laws, allowing regulators to see sensitive data without exposing vulnerabilities.

    The net effect is simple: companies can no longer cherry-pick what they share. Safety practices and risk events are on file, either for the public or for regulators.

    Unusually, SB 53 lets California’s Office of Emergency Services designate federal reporting standards as satisfying state law — even if federal rules don’t preempt. This creates a safe harbor: comply federally once, and you’ve complied in California. It’s a voluntary harmonization mechanism, and it signals California is eager for Washington to step up with national standards.

    Whistleblower Protections

    SB 53 prohibits frontier developers from preventing or retaliating against “covered employees” who disclose information about catastrophic risks or violations. The definition extends remarkably broadly: full-time employees, contractors, subcontractors, freelancers, unpaid advisors, vendors, and board members involved in assessing, managing, or addressing catastrophic risk.

    Companies must establish internal mechanisms allowing anonymous reporting and must provide updates on how concerns are addressed. Protected disclosures can go to the Attorney General, federal authorities, managers, or other employees with investigative authority if the employee has reasonable cause to believe information reveals specific and substantial danger to public health or safety from catastrophic risk, or that the developer violated SB 53.

    Why this matters: Internal employees often identify risks before external observers, but fear of retaliation prevents disclosure. SB 53’s broad definition and anti-retaliation provisions create formal channels for surfacing safety issues early. The inclusion of board members and unpaid advisors ensures virtually anyone with access to privileged risk information receives protection. This addresses concerns raised when former OpenAI employees criticized safety practices but faced potential legal exposure.

    Independent Audits: Delayed Accountability

    Starting in 2030, large frontier developers must engage independent third-party auditors annually to verify two elements: whether the developer follows their published frontier AI framework, and whether the framework is sufficiently clear that compliance can be objectively determined.

    This five-year delay gives the industry time to establish baseline practices before mandatory verification begins. The requirement addresses a fundamental challenge: voluntary commitments remain unverifiable without external oversight. By requiring auditors to assess both compliance and framework clarity, SB 53 creates pressure for concrete, measurable safety practices rather than aspirational principles.

    Why this matters: The delayed implementation distinguishes SB 53 from last year’s failed SB 1047, which mandated immediate annual audits. That prescriptive approach contributed to industry opposition and veto. The 2030 trigger represents legislative learning: establish transparency first, add verification once practices mature. But it also means five years of self-attestation before independent confirmation begins.

    Enforcement and Penalties

    The California Attorney General holds exclusive enforcement authority. Civil penalties are capped at $1 million per violation.

    The focus is not on punishing catastrophic harms themselves, but on penalizing failures of transparency — for example, not publishing a required framework, misrepresenting risk management, or failing to report incidents.

    SB 53’s liability structure reinforces its transparency-first philosophy. Companies that document their processes and report honestly cannot be sued under this statute even if their models cause catastrophic harm. But they do face significant penalties if they fail to live up to their own disclosures.

    CalCompute and State Preemption

    SB 53 establishes a consortium to develop CalCompute, a state-backed public cloud computing cluster providing free or low-cost resources to startups, researchers, and public-interest projects. The consortium must submit a framework by January 1, 2027, with preference for housing at the University of California. The initiative only proceeds if the state budget allocates funding.

    The law preempts local ordinances specifically related to frontier developer management of catastrophic risk adopted after January 1, 2025, preventing conflicting city and county regulations within California while leaving federal preemption unresolved.

    Practical Implementation Timeline

    SB 53 becomes effective January 1, 2026, creating a tight implementation window for covered developers:

    Before January 1, 2026:

    • Determine whether you meet both compute and revenue thresholds
    • Draft and publish frontier AI framework if covered
    • Establish catastrophic risk assessment processes
    • Create incident detection and reporting mechanisms
    • Implement whistleblower protections and anonymous reporting channels
    • Update employment agreements, contractor relationships, and board policies

    Ongoing Obligations:

    • Publish transparency reports before releasing new or modified models
    • Submit confidential risk assessments to OES for internal model use
    • Report critical safety incidents within required timelines
    • Update frontier AI framework annually
    • Maintain whistleblower investigation and response procedures

    Starting 2030:

    • Engage independent auditors annually to verify framework compliance and clarity

    Companies approaching thresholds should monitor growth trajectories. Revenue nearing $500 million or compute usage approaching 10^26 FLOPs triggers planning obligations before formal coverage begins.

    Looking Forward

    The law includes built-in adaptation mechanisms. The California Department of Technology must annually recommend updated definitions for “frontier model,” “frontier developer,” and “large frontier developer” based on multistakeholder input, technological developments, and international standards. This acknowledges that today’s thresholds might become obsolete as hardware improves and business models change.

    OES must publish annual anonymized incident reports starting in 2027, creating the first longitudinal dataset tracking frontier AI safety at state level. The Attorney General must similarly report on enforcement actions and whistleblower activity. These transparency mechanisms allow assessment of whether SB 53’s accountability measures produce meaningful behavior change.

    The 2030 independent audit requirement triggers the law’s most significant escalation. Once external verification begins, procedural compliance shifts from self-attestation to third-party review. However, technological development might outpace regulatory adaptation. If frontier capabilities advance faster than annual definition updates, dangerous systems might emerge outside SB 53’s scope.

    Conclusion

    SB 53 represents the United States’ first serious attempt to regulate frontier AI development through mandatory transparency requirements backed by civil penalties. Whether this approach succeeds depends on questions that cannot be answered today.

    California’s decision to regulate frontier AI safety marks a turning point in technology policy. For decades, Silicon Valley operated with minimal regulatory constraints, relying on voluntary industry commitments. SB 53 converts some voluntary commitments into binding legal obligations while deliberately avoiding the heavy-handed mandates that doomed last year’s attempt.

    The law embodies a particular philosophy about AI governance: trust but verify, transparency over prescription, accountability without stifling innovation. Whether that philosophy proves adequate for managing catastrophic risks from increasingly capable AI systems remains the defining policy question of our time.

  • Everything you need to know about GDPR [UPDATED: February 20, 2026]

    Everything you need to know about GDPR [UPDATED: February 20, 2026]

    This is educational material and does not constitute legal advice nor is any attorney/client relationship created with this article, hence you should contact and engage an attorney if you have any legal questions.


    GDPR Overview

    For seven years, the European Union’s General Data Protection Regulation (GDPR) has fundamentally reshaped how the world thinks about privacy, and completely broken websites the world over with its dreaded, mandated cookie banners.

    Since taking effect in May 2018, it has generated over €5.88 billion in fines, established Europe as the global privacy standard-setter, and forced every major technology company to rethink how they handle personal data.

    But GDPR is far more than just a European regulation. Its extraterritorial reach means that if your business processes data from EU residents — whether through a website, app, or any digital service — you’re likely subject to its requirements. And with AI and machine learning creating new compliance challenges, understanding GDPR has become essential for any business leader operating in the digital economy.

    However, GDPR’s complexity creates a compliance gap. Most small startups aren’t fully compliant; they’re either flying under the radar, doing bare minimum compliance, or completely ignoring it until they reach significant scale. But the stakes are real: GDPR violations can result in fines up to €20 million or 4% of global revenue, whichever is higher.

    This article breaks down what GDPR actually means for businesses, how enforcement has evolved since 2018, and why the regulation’s intersection with AI represents the next frontier of data protection compliance.

    The Foundation: What GDPR Actually Covers

    GDPR applies to the processing of personal data — essentially any information relating to an identified or identifiable person. This includes obvious categories like names and email addresses, but extends to IP addresses, location data, online identifiers, and even pseudonymized data that could reasonably be traced back to individuals.

    The regulation’s territorial scope is deliberately broad. It covers any organization established in the EU, regardless of where the data processing occurs. It also applies to organizations outside the EU that offer goods or services to EU residents or monitor their behavior. This “targeting criterion” has brought most global digital businesses within GDPR’s reach.

    Recent enforcement demonstrates this expansive scope. In 2024, Clearview AI received a €30.5 million fine from Dutch authorities despite being a US-based company with no EU operations. The company’s scraping of images from social media platforms and its facial recognition services were deemed to violate GDPR’s biometric data protections. The Dutch authorities didn’t care that Clearview argued it wasn’t targeting Europeans — the processing necessarily included EU residents — and therefore triggered GDPR obligations. This case illustrates how GDPR’s extraterritorial application continues to surprise companies that assumed they were beyond European regulatory reach.

    GDPR compliance hinges on five fundamental business considerations, but a company’s risk profile determines urgency:

    • High-Risk Companies (Immediate Priority):
    • B2B data processors handling other companies’ data
    • Financial services adjacent (insurance, fintech, etc.)
    • Companies with active EU expansion plans
    • Cross-border data flows with US vendors
    • Systematic monitoring or large-scale processing

    The five considerations to consider include:

    1. What types of personal data matter most for compliance risk? Basic identifiers like names and emails create standard obligations, but biometric data, location tracking, and behavioral profiling generate heightened scrutiny and larger fines. Meta’s €1.2 billion penalty for international data transfers and multiple sanctions for children’s data processing demonstrate how certain data categories attract disproportionate regulatory attention.
    2. Your EU footprint determines GDPR applicability through two key tests: Do you have any establishment in the EU (offices, employees, subsidiaries)? And do you process data from people located in the EU through your products or services? The regulation’s territorial scope is deliberately broad, covering any organization established in the EU regardless of where data processing occurs, plus organizations outside the EU that offer goods or services to EU residents or monitor their behavior. Even minimal EU connections can trigger full compliance obligations
    3. Your processing purposes and methods determine which legal requirements apply. Marketing communications typically require explicit consent, while service delivery can rely on contract performance. Operational analytics, e.g., like Google Analytics for instance, often use “legitimate interests” as a legal basis, but this requires balancing business needs against individual privacy rights. The choice affects everything from user experience design to vendor contracts.
    4. Third-party vendor relationships create some of GDPR’s biggest compliance challenges and enforcement risks. When you use cloud providers, analytics platforms, or marketing tools that process EU personal data, you remain liable for their compliance failures. Recent enforcement actions show regulators increasingly scrutinizing these processor relationships, particularly for international data transfers and security incidents.
    5. Finally, your security and privacy practices determine both your compliance posture and penalty exposure when incidents occur. GDPR requires “appropriate technical and organizational measures” including encryption, access controls, and regular security assessments. More critically, you need operational procedures for responding to individual rights requests within 30 days and breach notification to authorities within 72 hours. The regulation’s accountability principle means you must document these measures as regulators increasingly demand evidence of proactive compliance efforts when calculating fines.

    These categories might seem straightforward, but they determine both your GDPR obligations and potential liability exposure when things go wrong; we’ll unpack each of these in detail, below.

    Practical Implementation: A Phased Approach

    Effective GDPR compliance doesn’t happen overnight. Organizations should adopt a risk-based, phased strategy:

    Phase 1 (Immediate – 30 days):

    • Basic Data Processing Agreements with EU clients
    • Updated privacy policies
    • Standard Contractual Clauses with US vendors
    • Basic data mapping

    Phase 2 (Next 6 months):

    • Full Records of Processing Activities documentation
    • Data subject rights procedures (30-day response capability)
    • Comprehensive vendor audit

    Phase 3 (Growth stage):

    • Privacy by design processes
    • Regular compliance audits
    • Dedicated privacy resources

    GDPR Meets AI: The New Compliance Frontier

    The intersection of GDPR and artificial intelligence represents the regulation’s most dynamic and challenging application area. AI systems strain traditional privacy concepts around data minimization, purpose limitation, and individual rights while creating new categories of privacy risk.

    Training data compliance has emerged as a critical enforcement focus. The European Data Protection Board’s recent guidance clarifies that AI models cannot automatically be considered anonymous after training; rather, organizations must assess whether personal data can be extracted through various attack methods. If that sounds like trying to prove a negative, that’s precisely the seemingly insurmountable, Sisyphean challenge now facing AI-focused companies. That said, web scraping for AI training may rely on legitimate interests if organizations respect contextual privacy expectations and implement appropriate balancing assessments.

    Data minimization challenges create inherent tension between AI’s data requirements and GDPR’s necessity principles. Large language models require vast training datasets, while GDPR demands processing only data that’s adequate, relevant, and necessary for specific purposes. Organizations are exploring technical solutions including differential privacy, synthetic data generation, and federated learning (training models on local devices and then sharing only the resulting parameters) to address these tensions.

    Individual rights implementation becomes complex in AI contexts. Providing meaningful information about algorithmic logic (as required by Articles 13-15) proves difficult when the processing involves machine learning systems that even their creators don’t fully understand. Data portability requirements strain when personal data is embedded within trained models that can’t be easily extracted or transferred.

    The EU AI Act adds another layer of complexity, creating overlapping jurisdiction between data protection authorities and market surveillance bodies. Organizations developing or deploying AI systems must navigate both GDPR’s privacy requirements and the AI Act’s algorithmic governance obligations, requiring integrated compliance strategies that address both regulatory frameworks.

    Building Compliance Infrastructure

    Effective GDPR compliance requires more than privacy policies and cookie banners. Organizations need technical and organizational infrastructure capable of supporting ongoing data protection obligations.

    Data mapping forms the foundation of any compliance program. Organizations must understand what personal data they collect, where it comes from, how it’s processed, who has access, where it’s stored, and how long it’s retained. This visibility enables informed decisions about lawful bases, individual rights responses, and data protection impact assessments.

    Privacy by design integration embeds data protection considerations into system development and business processes from the outset. This includes implementing data minimization controls, automated retention schedules, access controls limiting data exposure, encryption for sensitive data, and technical measures supporting individual rights exercise.

    Vendor management requires comprehensive due diligence and ongoing oversight. Article 28 Data Processing Agreements must include specific provisions covering processing instructions, security measures, sub-processor authorization, data subject rights assistance, and data return or deletion obligations. Organizations remain liable for processor compliance failures, making vendor selection and monitoring critical compliance activities.

    Critical vendor categories requiring scrutiny:

    • Cloud storage (AWS, Google Cloud, Azure)
    • Analytics platforms (Google Analytics, Mixpanel)
    • Communication tools (Slack, email providers)
    • Payment processors
    • Marketing automation platforms

    Incident response capabilities must enable 72-hour breach notification to regulatory authorities and timely individual notification when appropriate. This requires detection capabilities, internal escalation procedures, impact assessment methodologies, and communication templates that can be rapidly deployed during security incidents.

    International Data Transfers: The Elephant in the Room

    If you’re a US-based startup with EU users — which, if you have a website, you probably are — international data transfers may be your single biggest practical GDPR compliance challenge. The regulatory landscape here has been turbulent, politically charged, and genuinely confusing. Here’s the state of play.

    The Schrems Saga. Austrian privacy activist Max Schrems has single-handedly dismantled two EU-US data transfer frameworks. In 2015, the European Court of Justice struck down the Safe Harbor framework (Schrems I) after Schrems argued that US surveillance practices were incompatible with EU fundamental rights. In 2020, the Court struck down Privacy Shield (Schrems II) on essentially the same grounds, sending thousands of companies scrambling to find alternative transfer mechanisms.

    The Current Framework: EU-US Data Privacy Framework (DPF). In July 2023, the European Commission adopted a new adequacy decision for the US based on the EU-US Data Privacy Framework. Under the DPF, US companies that self-certify through the Department of Commerce can lawfully receive EU personal data. This restored legal certainty — for now. But Schrems has already challenged the DPF (Schrems III is pending before the CJEU), arguing that the underlying US surveillance authorities haven’t materially changed. Many privacy professionals expect this framework to face serious judicial scrutiny.

    Standard Contractual Clauses (SCCs): The Workhorse. Regardless of whether the DPF survives, Standard Contractual Clauses remain the most widely used transfer mechanism. SCCs are pre-approved contractual terms issued by the European Commission that importers and exporters of personal data must sign. Since 2021, the Commission has required the updated “modular” SCCs, which replaced the older versions. If you’re a US-based SaaS company processing EU data, you should have SCCs in place with your customers or incorporate them by reference in your Terms of Service or Data Processing Agreement.

    Critically, SCCs aren’t a set-and-forget solution. Post-Schrems II, the EDPB requires organizations to conduct a Transfer Impact Assessment (TIA) evaluating whether the laws of the importing country (here, the US) provide essentially equivalent protection to EU law. For most low-risk SaaS applications, this assessment is relatively straightforward — but you need to document it.

    The CLOUD Act Concern. The US CLOUD Act (Clarifying Lawful Overseas Use of Data Act, 2018) allows US law enforcement to compel US-based companies to produce data regardless of where that data is physically stored. This is one of the underlying tensions that drove the Schrems II decision, and it remains a live concern among EU privacy advocates and compliance officers.

    However — and this is important — the CLOUD Act does not make it illegal for EU residents to use US-hosted services. It creates a theoretical risk of compelled disclosure, but that risk must be evaluated in context. For a SaaS subscription tracker storing app names and monthly costs, the realistic probability of a US law enforcement demand is effectively zero. For a healthcare platform storing patient records, the calculus is very different. This is why the TIA matters: it forces a risk-proportionate assessment rather than a blanket prohibition.

    What US Startups Should Actually Do. If you process EU personal data, your practical compliance checklist for international transfers looks like this:

    1. Reference SCCs in your privacy policy and/or Data Processing Agreement as your primary transfer mechanism.
    2. Check whether your sub-processors are DPF-certified. Major providers like Stripe, Google, and AWS are. If they are, cite the DPF as an additional safeguard.
    3. Conduct and document a Transfer Impact Assessment, even a lightweight one. For most B2B SaaS applications processing low-sensitivity data, this can be a 2-3 page document evaluating the nature of the data, the likelihood of government access requests, and the supplementary measures you’ve implemented (encryption, access controls, etc.).
    4. Don’t rely on consent as your transfer mechanism. GDPR requires that consent for international transfers be explicit, specific, informed, and freely given. “By using this service you consent to processing in the US” buried in a privacy policy does not meet this standard. Use SCCs instead.
    5. Consider EU-hosted infrastructure if you’re seriously targeting the EU market. It’s not legally required if you have proper transfer mechanisms, but it eliminates the objection entirely and can be a competitive advantage with privacy-conscious European customers.

    The international transfer regime is likely to keep evolving. If Schrems III invalidates the DPF — which is a real possibility — SCCs will again become the sole viable mechanism for most companies, and the pressure to implement supplementary technical measures (like EU-based hosting or end-to-end encryption) will increase significantly. Building your compliance posture around SCCs rather than the DPF alone is the more durable strategy.

    Looking Forward: GDPR’s Evolving Landscape

    GDPR continues evolving through judicial interpretation, enforcement action, and technological change. Recent European Court of Justice decisions have clarified key concepts around pseudonymization, damages thresholds, and international transfers, while data protection authorities develop technical expertise in emerging areas like AI and biometric processing.

    Regulatory convergence reflects GDPR’s global influence, with privacy laws in Brazil, India, California, and numerous other jurisdictions incorporating similar principles and requirements. However, geopolitical fragmentation through data localization mandates, national security restrictions, and laws like the US CLOUD Act may challenge unified global approaches to data governance. The tension between US government access authorities and EU fundamental rights protections remains unresolved and will likely drive continued litigation and regulatory evolution around international data transfers.

    The relationship between GDPR and the EU AI Act will shape future compliance requirements for organizations deploying AI systems. Both regulations emphasize risk-based approaches, transparency obligations, and individual rights, but their interaction creates complex overlapping requirements that organizations must navigate carefully.

    Enforcement sophistication will continue increasing as regulators develop technical expertise and coordinate cross-border investigations. Organizations should expect more targeted enforcement actions focusing on emerging technologies, children’s privacy, and international data transfers as regulatory authorities mature their enforcement capabilities.

    For business leaders, GDPR represents both compliance obligation and competitive opportunity. The regulation’s complexity creates a compliance gap, particularly for resource-constrained organizations, but certain factors — like processing other companies’ data, regulated industry exposure, or active EU market targeting — make compliance non-optional. Organizations that invest in sophisticated privacy governance capabilities while embracing privacy-enhancing technologies will be best positioned for success in an increasingly regulated digital economy. The key is adopting a risk-based, phased approach that matches compliance investment to actual exposure.

    For further questions, feel free to reach out.

  • ⚠️ EU AI Act enforcement penalties go into effect from Sunday, Feb 2, 2025

    ⚠️ EU AI Act enforcement penalties go into effect from Sunday, Feb 2, 2025

    This is educational material and does not constitute legal advice nor is any attorney/client relationship created with this article, hence you should contact and engage an attorney if you have any legal questions. No warranties, express or implied, are made with respect to its accuracy. Information contained herein, or information relied upon, is subject to change without notice.


    Enforcement penalties under Article 5 of the EU AI Act (“EUAIA”) will go into effect on February 2, 2025.This significant step aims to ensure AI technologies are developed and used in a manner that is safe, transparent, and respects fundamental rights. For a deep dive into the EUAIA, you can read more here.

    Prohibited AI Practices

    From February 2, 2025, certain AI applications will be outright banned due to their unacceptable risk levels. These include manipulative AI systems, social scoring, and emotion recognition systems in certain contexts like workplaces and schools. The penalties for non-compliance could be severe, reaching up to €35 million or 7% of the global annual turnover for the entity, whichever is higher.

    AI Literacy

    A notable aspect of the EUAIA is the mandate for “AI literacy”. Per Article 4 of the EUAIA, providers and deployers of AI systems must ensure their users are equipped with a sufficient understanding of AI. This requirement aims to foster responsible use, although the direct penalties for failing to meet this standard are not explicitly detailed in the initial phase.

    Enforcement and Compliance

    Companies operating within or targeting the EU market must prepare for compliance, as the Act applies to both providers and users of AI systems. This includes ensuring that AI systems do not infringe on the prohibitions set forth by the EUAIA.

    For examples of prohibited AI use,

    For a substantive deep dive into the EUAIA, you can learn more here.

  • Everything you need to know about the EU AI Act (Updated April 14, 2025)

    Everything you need to know about the EU AI Act (Updated April 14, 2025)

    This is educational material and does not constitute legal advice nor is any attorney/client relationship created with this article, hence you should contact and engage an attorney if you have any legal questions.


    We should begin this discussion with an important clarification:

    The European Union AI Act was not in fact conceived in response to the new era of so-called “generative AI” apps that began on November 30, 2022, when ChatGPT launched for public use. On the contrary, the first iteration of the EU AI Act (“EUAIA”) was drafted in April 2021 to regulate artificial intelligence systems generally, not “generative AI” systems like ChatGPT specifically.

    Indeed, it was only subsequent revisions to the EUAIA that sought to incorporate text to accommodate generative AI following the launch of ChatGPT and others. Hence, the final version we study here is essentially an afterthought to compensate for what had been a decidedly unforeseen advent in the evolution of artificial intelligence systems. Simply put, nobody had anticipated generative AI specifically, with its almost mysteriously complex large language models (“LLMs”; learn more here) — at least not any time soon — and so the drafters of the EUAIA had to adapt quickly.

    The final result, published in the Official Journal of the European Union on July 12, 2024, and entering into force on August 1, 2024, is a landmark regulatory framework designed to ensure AI technologies are developed, deployed, and used responsibly. While critics may well fault it for its excessively strong grip akin to the EU’s GDPR (General Data Protection Regulation), its intent is unquestionably sound and indeed necessary to ensure safety in our fast-changing digital world, where fantastical things once limited to the confines of science fiction are now very real indeed.

    This article attempts to unpack and summarize the nuances of the EUAIA, as derived directly from the EUAIA’s entire text and official summary, including key definitions, risk classifications, compliance requirements, penalties, and the implications for AI providers and deployers. A “mind map” is available below for an even deeper dive for those so inclined to learn more.

    EU AI Act Mind Map Preview Image
    EU AI Act Mind Map. Copyright Hoag Law.ai. All Rights Reserved. Contact to request high-res PDF.

    Key Definitions and Scope

    The EUAIA defines an AI system as any machine-based setup that makes inferences to achieve objectives, except systems for military or defense purposes. It uses a functional, effect-based definition, highlighting characteristics such as autonomy, adaptability, and data dependence. The EUAIA applies to providers placing AI systems on the EU market, as well as to providers and deployers of AI systems in third countries if the output is used in the EU. But the EUAIA includes exceptions for free and open-source AI systems, unless they are otherwise prohibited or classified as high-risk AI systems, as defined below.

    Risk Tiers for AI Systems

    EU AI Act Risk Tiers Pyramid
    EU AI Act Risk Tiers. Copyright © Hoag Law.ai. All Rights Reserved.


    The framework of the EUAIA is premised on a categorization of AI systems into three primary risk tiers: prohibited, high-risk, and limited-risk. General-purpose and minimal-risk tiers are also slotted into the hierarchy.

    Overall, the EUAIA mandates transparency for all AI systems, especially those that interact with humans or are involved in emotion recognition or biometric categorization; meanwhile, so-called high-risk AI systems must carry clear labels and instructions.

    Providers of general-purpose AI models are required to disclose detailed technical documentation, including information about training data, limitations, and capabilities. Free and open-source GPAI models are exempt from some requirements unless they present systemic risk.

    The EUAIA also includes specific provisions for deep fakes, defined as AI-generated or manipulated content that falsely appears authentic. Providers and deployers of such systems must disclose that the content is artificially generated or manipulated, with certain exceptions for authorized law enforcement use or evidently artistic works.

    To foster innovation, the EUAIA includes provisions for AI regulatory sandboxes, allowing for supervised testing of novel AI applications. It also provides support measures for SMEs and startups to help them navigate compliance requirements.

    We shall now unpack each of these in decreasing order from highest to least risk.

    Prohibited AI Practices

    The EUAIA bans certain AI practices outright due to their potential for harm. These include:

    • Manipulative AI that distorts human behavior.
    • Social scoring systems that may lead to discriminatory outcomes.
    • Real-time remote biometric identification by law enforcement, except in cases of severe public interest.
    • Biometric categorization systems inferring sensitive attributes, with specific exceptions for law enforcement use.

    High-Risk AI Systems

    High-risk AI systems are those with significant potential to impact health, safety, or fundamental rights. These systems must adhere to stringent requirements, including continuous risk assessment, high-quality data governance, comprehensive technical documentation including labeling requirements, robust quality management systems, and detailed record-keeping. The EUAIA outlines specific high-risk categories in Annex III of the EUAIA, including:

    • Biometric identification and categorization.
    • Critical infrastructure management.
    • Education and vocational training.
    • Employment, workers’ management, and access to self-employment.
    • Essential private and public services.
    • Law enforcement.
    • Migration, asylum, and border control management.
    • Administration of justice and democratic processes.

    General Purpose AI Systems

    These are AI models designed to perform a wide range of tasks and are assumed with systems presenting systemic risks (e.g.,those that exceed 10^25 FLOPs (floating point operations per second; for more information, see here)). Such systems have specific obligations under the EUAIA, including:

    • Providing technical documentation and instructions for use: Publishing a “training data transparency summary” that includes the datasets used, their collection methods (e.g., web scraping or third-party sources), licensing status, legal basis, and whether copyright opt-outs were honored.
    • Complying with EU copyright law.
    • Publishing a summary of training data content.
    • Free and open-source GPAI models are exempt from some requirements unless they present systemic risk.

    The EUAIA also places specific obligations on providers of general-purpose AI models regarding copyright compliance and transparency about training data:

    • Providers must implement a policy to comply with EU copyright law: A text and data mining exception allows reproductions and extractions of lawfully accessible works, unless rightsholders opt out via machine-readable means (e.g., metadata, robots.txt, ai.txt, or T&Cs).
    • The EUAIA recognizes a text and data mining exception, allowing for reproductions and extractions of lawfully accessible works for AI training purposes, but rightsholders can opt out, requiring AI providers to obtain authorization for text and data mining of their works.
    • Providers must publish a sufficiently detailed summary of content used for training, facilitating copyright enforcement.
    • These obligations apply to all GPAI models on the EU market, regardless of where the training occurred.

    Limited Risk AI Systems

    Limited risk systems are those that are not high-risk yet still require transparency and accountability. Somewhat surprisingly, deep fakes are included in this risk category, despite their profound potential to disseminate false or misleading information by impersonating people without their consent. Providers must ensure users are aware they are interacting with an AI system and maintain basic documentation.

    Minimal-Risk AI Systems

    This category encompasses the majority of AI systems currently on the market. These systems:

    • Are not subject to specific obligations under the EUAIA due to their low risk profile.
    • Are encouraged to follow voluntary codes of conduct to ensure ethical and responsible AI development and use.
    • Allow for continued innovation while promoting best practices in AI governance.

    Compliance and Penalties

    Compliance with the EU AI Act involves a multi-faceted approach:

    • Conformity Assessments: High-risk AI systems must undergo pre-market evaluations.
    • Post-Market Monitoring: Continuous surveillance to ensure ongoing compliance.
    • Market Surveillance Authorities: National bodies responsible for enforcement.
    • EU Database: A centralized repository for high-risk AI systems.
    • Notified Bodies: Independent entities conducting conformity assessments.

    Penalties for non-compliance are broken down as follows:

    • Prohibited uses can incur fines up to €35 million or 7% of global annual turnover, whichever is higher and/or prohibition from use.
    • High-risk compliance failures can attract fines up to €15 million or 3% of global turnover, whichever is higher and/or full access to system data and source code for inspection.
    • Misleading information can result in fines up to €7.5 million or 1% of global turnover, whichever is higher and/or proactive market surveillance.

    For SMEs and startups, the fines are subject to the same maximum percentages or amounts, whichever is lower.

    Deployers’ Obligations

    Deployers of AI systems have specific obligations under the EUAIA. They must implement a risk assessment system, adopt transparency principles, and design responsible AI governance frameworks. Additionally, deployers of high-risk AI systems in public services must conduct fundamental rights impact assessments to ensure their systems do not negatively impact rights.

    Data Protection and Privacy

    Alignment with the General Data Protection Regulation (GDPR) is a cornerstone of the EUAIA, ensuring that AI systems comply with stringent data protection laws. Specific rules govern the processing of biometric data, imposing strict conditions to safeguard privacy.

    Governance and International Cooperation

    The governance structure established by the EUAIA includes the European Artificial Intelligence Board, which coordinates regulatory practices across Member States, and the AI Office, which oversees AI regulations and offers guidance. A scientific panel of independent experts provides technical and ethical advice. Internationally, the Act promotes global cooperation to ensure that AI systems developed outside the EU comply with EU standards. Mutual recognition agreements facilitate conformity assessments with other jurisdictions, ensuring a global standard for AI safety and compliance.

    Implementation Timeline

    The EUAIA’s implementation is slated to be phased in accordingly, with key dates including:

    • August 1, 2024: The EUAIA officially goes into effect.
    • February 2, 2025: Prohibited uses become enforceable.
    • May 2, 2025: Codes of practice for general-purpose AI models are made available. (This will be interesting to see.)
    • August 2, 2025: Compliance deadline for general-purpose AI providers.
    • February 2, 2026: The Commission provides guidelines and practical examples. (Also this.)
    • August 2, 2026: Full applicability of the EUAIA, with operational infrastructure.
    • August 2, 2027: Extended compliance for certain high-risk AI systems.

    Ongoing Case Law and Compensation Frameworks

    In September 2024, the Hamburg Regional Court ruled in favor of LAION, a non-profit that distributed datasets with links to copyrighted content, citing a text and data mining exception for scientific purposes.

    In March 2025, French publisher and author organizations sued Meta Platforms Inc. over alleged unauthorized use of copyrighted works to train Llama, a generative AI model.

    French policymakers are also exploring collective rights management and trusted intermediaries to ensure authors are compensated fairly when their content is used in AI training.

    Conclusion

    Like it or not, the EU AI Act is an impressively comprehensive regulatory framework designed to manage the rapid evolution of artificial intelligence, including and especially generative AI with all of its controversial copyright infringement allegations for language model (LLM) training data. By establishing robust compliance mechanisms, promoting transparency, and fostering innovation, the EUAIA aims to create a balanced ecosystem where AI can thrive while safeguarding fundamental rights and public interest, even if it may seem stifling. For AI developers, providers, and deployers, understanding and adhering to the EU AI Act is crucial for navigating the new regulatory landscape effectively.

    For further questions, feel free to reach out.

  • Using AI in Law: Compliance with ABA Model Rules and Best Practices [UPDATED: February 19, 2026]

    Using AI in Law: Compliance with ABA Model Rules and Best Practices [UPDATED: February 19, 2026]

    This is educational material and does not constitute legal advice nor is any attorney/client relationship created with this article, hence you should contact and engage an attorney if you have any legal questions. Any discussion on technical matters may be inaccurate or no longer timely, so you should consult with the various vendors mentioned below to clarify and to ascertain accuracy.


    The ABA Model Rules of Professional Conduct strongly imply, if not yet require, the adoption of generative AI technology in the practice of law.


    The legal sector is rapidly embracing generative AI to streamline tasks such as research, drafting, document review, finding the needle in a haystack of email discovery, and more.

    The case in 2023 of a New York lawyer being sanctioned for filing a brief with fake AI-generated case citations first alerted the world with a cautionary reminder: while AI can enhance our capabilities, it must be wielded with utmost responsibility to prevent ethical violations and safeguard our reputation.

    While hallucinations have largely vanished as of this update in February 2026, attorneys must always review AI-generated output just as they review other associates’ and paralegals’ output, too.

    Accordingly, the ABA Model Rules of Professional Conduct and California Rules of Professional Conduct, which largely align with the ABA Rules, not only imply but arguably necessitate the adoption of technology, including generative AI, to maintain competent client representation.

    Embracing Competence through Technology

    Maintaining competence is a fundamental ethical obligation for lawyers. According to ABA Rule 1.1 on Competence, this includes staying updated on changes in the law and its practice, including the benefits and risks associated with relevant technology (Comment 8 to ABA Rule 1.1).

    This rule arguably implies that lawyers should use AI tools to enhance their practice, while emphasizing the need to ensure the accuracy and reliability of AI-generated content.

    Regular training and the adoption of advanced AI tools are prudent ways to fulfill this obligation, ensuring that lawyers are well-versed in both the capabilities and limitations of AI technologies.

    Enhancing Diligence with AI

    ABA Rule 1.3 on Diligence requires lawyers to act with reasonable diligence and promptness in representing a client; this therefore implies strongly that technology must be used to balance a lawyer’s workload.

    A lawyer relying on pen and paper to draft in days or weeks what would ordinarily take hours with a word processor, or to eschew email entirely for communicating with opposing counsel in favor of using only postal mail, would arguably run afoul of this rule if only because of the unnecessary delays and resulting costs that such anachronistic work would impose upon clients.

    AI tools can likewise be a significant asset in managing workloads and delivering timely service to clients. By automating routine tasks like legal research, document review, drafting, and sifting through hundreds and thousands of discovery documents, lawyers can free up valuable time for more complex and strategic aspects of their work; in some instances the time and cost savings can be profound, sometimes reducing to mere minutes what would have otherwise taken hours.

    However, it is crucial to remember that AI is not a substitute for human judgment and expertise, and errors are still a concern. Lawyers must still exercise diligence in reviewing and verifying AI-generated content to ensure its accuracy and relevance to the specific case at hand.

    Ensuring Accuracy in AI Outputs

    ABA Rule 3.1 on Meritorious Claims requires that lawyers only bring forward claims that are based on law and fact. Comment 2 to this rule emphasizes that an “action is frivolous … if the lawyer cannot make a good faith argument on the merits of the action taken” or support it by a good faith argument for the extension, modification, or reversal of existing law.

    Therefore, it is essential to independently verify the accuracy of AI-generated content, including sources, citations, case law, statutes, and quoted language.

    Manually checking all citations and legal references against primary legal sources is crucial, and simply asking or “prompting” a generative AI tool to “double-check its work,” as it were, will not usually suffice.

    What can sometimes help is to use multiple generative AI platforms in parallel, by running your desired output from the first platform back and forth through all of them, as if slowly distilling the end result through a funnel to filter out the errors.

    Maintaining Transparency and Protecting Confidentiality

    Transparency in disclosing AI usage not only fosters trust but also helps the court and clients understand the role of AI in your work. ABA Rule 1.4 on Communication emphasizes the importance of keeping clients reasonably informed about the status of their matter.

    When using AI tools, lawyers should disclose this to their clients and explain how AI is being used to assist with their case. This can be done through engagement letters or other client communications. Below is an example of such a clause:

    Use of Artificial Intelligence (AI)

    In representing you, I may use generative AI tools for tasks such as legal research, drafting, email and file management, and analysis. These tools augment, but do not replace, my professional judgment. Any use of such tools will be consistent with applicable ethics rules, and I will maintain human oversight and protect your confidential information.

    However, ABA Rule 1.6 on Confidentiality mandates that lawyers must protect client information. When using AI tools, particularly those connected to public APIs, it is crucial to ensure that no confidential information is inadvertently disclosed.

    Crucially, when evaluating any AI platform for use with client-confidential information, attorneys should assess the following factors — roughly in decreasing order of importance:

    1. Zero Data Retention (ZDR) Policies. The single most critical requirement. Confirm that the AI provider contractually commits to not retaining your prompts, outputs, or uploaded documents beyond the duration of the session or API call. This is distinct from merely “opting out” of training — ZDR means the provider does not store your data at all, period. Major providers like OpenAI and Anthropic offer ZDR through their API and enterprise tiers if negotiated, but their free consumer products typically do not provide this protection. Always verify ZDR commitments in the provider’s Data Processing Agreement (DPA) or enterprise terms, not just their marketing materials.
    2. No Training on Your Data. Separately from retention, confirm that the provider does not use your inputs or outputs to train, fine-tune, or improve its models. While most enterprise and API tiers now exclude customer data from training by default, consumer-tier products often do not. This should be an explicit contractual commitment, not merely a toggle in a settings menu that can be changed unilaterally.
    3. Data Isolation and Access Controls. Understand the provider’s architecture: Is your data logically or physically isolated from other customers? Who at the provider has access to your data, and under what circumstances (e.g., trust and safety review, debugging)? Enterprise deployments and private instances offer stronger isolation, but even shared-infrastructure API access can be acceptable if ZDR and no-training commitments are in place.
    4. Encryption in Transit and at Rest. Ensure the platform uses industry-standard encryption (TLS 1.2+ in transit; AES-256 at rest). This is table stakes — virtually every reputable provider meets this standard — but it should still be confirmed, particularly for any data that is retained (e.g., conversation logs you choose to save on the platform).

    The critical distinction is between cloud storage (where the platform functions like a filing cabinet, holding your data for your later retrieval) and data processing (where the platform processes your data ephemerally and discards it).

    For AI tools handling privileged information, you want processing without storage — and ZDR is how you get there.

    And no, public AI platforms are not the same as cloud storage. The former is considered disclosure to a third party and thus risks waiving privilege, while the latter is not.

    Finally, if the platform allows you to save conversation histories or upload documents for persistent use, confirm that (a) such data is deletable on demand, (b) deletion is actual deletion and not merely soft-deletion or archival, and (c) you understand the provider’s data retention schedule for any backups or logs.

    My AI Privacy Guide was built for precisely this reason: to help you understand the various platforms’ Terms of Service and Privacy policies so that you can make a better informed decision.

    Ethical Supervision of AI Tools

    As AI tools — generative AI specifically — become more prevalent in legal practice, lawyers must remember that they are ultimately responsible for the work product generated using these tools.

    ABA Rules 5.1 and 5.3 emphasize the ethical duties of lawyers to supervise subordinate lawyers and non-lawyer assistants, including AI tools. This involves understanding how the AI tool works, monitoring its output for accuracy and reliability, and ensuring that its use aligns with ethical and professional standards, including and especially, matters involving privacy and confidentiality.

    Addressing Emerging AI Challenges

    The rapid advancement of AI in the legal industry raises novel ethical and legal questions that lawyers must be prepared to address. ABA Resolution 112 (2019) urges courts and lawyers to address emerging issues related to AI usage, such as algorithmic bias, explainability, and transparency.

    By staying informed about these issues and proactively implementing best practices for ethical AI usage, lawyers can harness the power of AI to improve efficiency and client service while upholding the highest standards of professionalism and integrity. This proactive approach includes engaging in continuous education about AI technologies and their ethical implications.

    If you’re curious to try an interesting academic thought exercise, consider asking your favorite AI chatbot — be it ChatGPT; Claude; or GC.AI — to draft a new set of fictitious rules for the ABA or California Rules of Professional Conduct governing the use of AI. I used the prompt below, but you should play around and see what you come up with:

    Can you please produce a fictitious rule for the California Rules of Professional Conduct, about the use of AI like LLMs like ChatGPT? Please give it appropriate headings and section numbers to fit into a logically coherent place in the existing rules, make sure it’s suitably detailed, and so on.

    I don’t want to give anything away, but I think you’ll find this exercise as entertaining as it is awe-inspiring and impressive. And while it obviously offers no practical value, it certainly suggests a very plausible direction to where things are likely headed with generative AI and the legal community.

    Spoiler alert: I’ll state for the record that I think it will soon be ethical malpractice NOT to use AI in the practice of law, or indeed, any profession, very soon; say, 2030.

    Closing Thoughts

    Generative AI is a revolutionary, society-changing tool the likes of which humanity has never seen before; a fantastic magic heretofore relegated only to the realm of fantasy and science fiction. It can significantly enhance the efficiency, accuracy, and client service in legal practice, but is not yet a replacement for human judgment and expertise.

    Attorneys must remain ethically responsible for their work product, regardless of the tools used. By selecting appropriate AI tools, verifying AI-generated content, adhering to court rules, transparently disclosing AI usage, and addressing the ethical implications of AI, lawyers can effectively integrate AI into their practice while avoiding potential pitfalls.

    And despite the risks, it is precisely because of the immense power and productivity gains offered by generative AI that the ABA and California Ethics Rules at least impliedly support and potentially encourage the thoughtful use of such technology to uphold the high standards of competence and ethical practice required in the practice of law, which at the end of the day, must always be in the client’s best interest.

    For more insights on responsible AI usage in legal practice, feel free to reach out or explore our resources.

  • Everything you need to know about Regulation Crowdfunding (“Reg CF”)

    Everything you need to know about Regulation Crowdfunding (“Reg CF”)

    This is educational material and does not constitute legal advice nor is any attorney/client relationship created with this article, hence you should contact and engage an attorney if you have any legal questions.


    Regulation Crowdfunding (“Reg CF”) has revolutionized how startups and small businesses raise equity capital, enabling them to secure up to $5 million in a 12-month period from a large number of non-accredited investors.


    Overview of Regulation Crowdfunding

    The Jumpstart Our Business Startups (JOBS) Act, signed into law on April 5, 2012, included Title III, known as the “Capital Raising Online While Deterring Fraud and Unethical Non-Disclosure Act of 2012” or the “CROWDFUND Act.” Title III created a new exemption under Section 4(a)(6) of the Securities Act of 1933, codified as 15 U.S.C. § 77d(a)(6), allowing eligible companies to raise funds through crowdfunding without registering the offering with the SEC, provided certain conditions are met.

    To implement this exemption and establish the regulatory framework, the Securities and Exchange Commission (SEC) adopted Regulation Crowdfunding (17 CFR Part 227) on October 30, 2015, with the regulations becoming effective on May 16, 2016. Reg CF allows eligible companies to offer and sell securities to a broad investor base, including non-accredited investors, through online crowdfunding platforms, subject to specific regulatory requirements and limitations.

    If “crowdfunding” sounds familiar, that’s because you’ve likely heard of platforms such as Kickstarter or Indiegogo, among others. While these platforms enable non-accredited investors to contribute money, the contributions are more akin to pre-purchases or gifts; they money put into these companies are not investments.

    In contrast, Reg CF platforms like WeFunder, StartEngine, Dealmaker Securities, and others, enable non-accredited investors to actually invest in startups, with all the associated risks and rewards, thereby democratizing the investment landscape by letting new startups’ biggest fans support the things they care most about.

    This comprehensive guide provides a detailed overview of Reg CF, its key requirements, limitations, and best practices for issuers and intermediaries.

    Reg CF Mind Map. Email me to get a high-res PDF.

    The Spectrum of Fundraising Regulations

    Startups and small businesses often follow a progression of fundraising stages, each with its own set of regulations and requirements. Understanding this spectrum can help companies strategically plan their capital-raising efforts while considering the cost, time, and pros and cons of each option.

    • Bootstrapping involves using personal savings and reinvested revenue to fund the business. This method offers full control and no dilution of ownership, but it limits the amount of available capital and increases personal financial risk. It is a quick method to start with low external costs.
    • Friends and Family funding involves raising initial funds from personal networks. It offers easier access to capital with flexible terms, though it can strain personal relationships and is limited in scale. This method is relatively quick and low-cost, though often informal.
    • Angel Investors provide capital in exchange for equity. They offer larger capital amounts and can provide mentorship and networking opportunities. However, this approach dilutes ownership and often requires giving up significant equity. Securing angel investment takes moderate time and incurs costs associated with legal and due diligence processes.
    • Reg CF (Regulation Crowdfunding) allows raising up to $5 million from a wide range of investors, including non-accredited investors, through online crowdfunding platforms. This method democratizes access to capital and offers marketing benefits but requires compliance with regulatory standards, disclosure requirements, and ongoing reporting. The preparation and compliance can take a moderate to long duration, and costs include fees to access the crowdfunding platform, or portal, and compliance-related expenses, too.
    • Venture Capital (Reg D 506(b)) involves attracting larger investments from venture capital firms through private placements, overwhelmingly under SEC Rule 506(b). Reg D 506(b) rounds are private offerings exempt from SEC registration, requiring minimal disclosures, and allowing companies to raise substantial capital without going public. This approach provides strategic support from experienced investors but results in significant dilution of ownership and high expectations for growth. The process is long and competitive, with significant legal and compliance costs.

    Key Definitions

    To better understand Regulation Crowdfunding, it’s essential to familiarize yourself with some key terms:

    • Issuer: A company offering and selling securities through Regulation Crowdfunding.
    • Intermediary: A crowdfunding platform, either a registered broker-dealer or a funding portal, that facilitates Reg CF transactions.
    • Investor: An individual or entity that purchases securities offered through a Reg CF offering.
    • Accredited Investor: An individual with an annual income exceeding $200,000 ($300,000 with spouse) or a net worth over $1 million, excluding the primary residence.
    • Crowdfunding: The practice of raising small amounts of money from a large number of people, typically via the internet.

    Eligibility Requirements and Offering Limits

    Startup (“issuer”) Eligibility

    To be eligible for a Reg CF offering, an issuer must:

    • Be U.S.-based: The entity must have its headquarters and principal place of business in the United States. This does not entirely preclude foreign entities from utilizing Reg CF, but they must establish a U.S.-based headquarters and do actual business in the U.S., and not merely funnel revenue to a foreign office.
    • Not be an investment or shell company: The issuer must not be an investment company, blank check company, or shell company as defined under the Securities Exchange Act of 1934. This means that the company must have a well-defined, operational business plan and cannot exist solely to acquire or merge with other companies.
    • Not be disqualified: The issuer must not be disqualified under Reg CF’s “bad actor” provisions, which include disqualifications due to certain criminal convictions, regulatory or court orders, and other legal sanctions.

    Offering Limits

    Reg CF allows issuers to raise up to $5 million in a 12-month period. The offering limit is determined by aggregating all Reg CF offerings conducted by the issuer and its affiliates within the preceding 12 months including the current raise. This means that if an issuer is attempting to raise $3M today, but they already raised $3M less than 12 months ago, then they will be limited to raising just $2M for the current raise.

    Investment Limits for Non-Accredited Investors

    Non-accredited investors are subject to investment limits based on their annual income and net worth:

    • If either annual income or net worth is less than $124,000, the investor may invest the greater of (a) $2,500 or (b) 5% of the greater of their annual income or net worth.
    • If both annual income and net worth are equal to or more than $124,000, the investor may invest up to the greater of: (a) 10% of their annual income, or (b) 10% of their net worth, not to exceed $124,000.

    Insider Transactions and Bad Actor Disqualification

    Insider Transactions

    Insiders, such as officers, directors, and significant shareholders, are required to disclose their participation in a Reg CF offering. Transactions involving insiders are subject to additional scrutiny to prevent conflicts of interest and self-dealing.

    Bad Actor Disqualification

    Issuers are disqualified from conducting a Reg CF offering if they or certain related parties (e.g., officers, directors, significant shareholders) have been involved in specified bad acts, such as securities fraud or other regulatory violations.

    Disclosure Requirements and Offering Process

    Disclosure Requirements

    Issuers must provide comprehensive disclosures to potential investors, including:

    • Information about the issuer, its officers, directors, and significant shareholders.
    • Description of the business and its financial condition, including financial statements.
    • Intended use of proceeds and target offering amount.
    • Offering price and valuation of securities.
    • Ownership and capital structure.
    • Risk factors associated with the investment.

    Financial Disclosure Tiers

    The extent of financial disclosures required depends on the amount being raised plus whatever has been raised in a Reg CF round in the past 12 months.

    • Tier 1 (offerings up to $124,000): Certified income tax returns and financial statements certified by the principal executive officer. The financial statements must cover the most recently completed fiscal year.
    • Tier 2 (offerings between $124,000 and $618,000): Financial statements reviewed by an independent public accountant. The financial statements must cover the most recently completed fiscal year.
    • Tier 3 (offerings more than $618,000 but not more than $1.235 million): Financial statements reviewed by an independent public accountant. If audited financial statements are available, they must be provided instead. The financial statements must cover the two most recently completed fiscal years or the period since inception, if shorter.
    • Tier 4 (offerings more than $1.235 million, up to $5 million): Financial statements audited by an independent public accountant. The financial statements must cover the two most recently completed fiscal years or the period since inception, if shorter.

    The specific disclosure requirements change somewhat depending on whether the startup was incorporated less than 120 days ago; more than 120 days ago but within the same year; or more than one year ago. The following matrix should help clarify the subtleties:

    Note that GAAP financials require a cover page; balance sheet; income statement; cash flow statement; stockholder equity statement; and footnotes which spell out details about taxes, equity, debt, accounting methodologies, or anything else of substantive or material importance; the idea here is transparency.

    Filing Requirements and Offering Process

    Issuers must file Form C with the SEC and provide it to the intermediary before commencing the offering. The offering must be conducted exclusively through a single intermediary platform. The offering must be open for at least 21 days before any securities can be sold. Investors can cancel their commitments for any reason up to 48 hours before the offering deadline.

    Intermediary Requirements and Funding Portals

    Intermediary Obligations

    Intermediaries facilitating Reg CF offerings must:

    • Register with the SEC and FINRA as either a broker-dealer or funding portal.
    • Conduct due diligence on issuers and screen for potential fraud.
    • Provide investor education materials and ensure investor acknowledgment of risks.
    • Facilitate the offering and closing process, including the transfer of funds.

    Funding Portal Regulations

    Funding portals are subject to specific regulations, including:

    • Prohibition on offering investment advice or recommendations.
    • Restrictions on soliciting investments and handling investor funds.
    • Requirements for recordkeeping and compliance with SEC and FINRA (Financial Industry Regulatory Authority) rules.

    Ongoing Reporting and Compliance

    Annual Reporting Requirements

    Issuers must file an annual report, known as Form C-AR, with the SEC and provide it to investors within 120 days of the end of the fiscal year. The annual report must include updated financial statements and information about the issuer’s business operations and financial condition. This ongoing reporting ensures transparency and accountability to investors.

    Terminating Reporting Obligations

    Issuers may terminate their ongoing reporting obligations under Reg CF if:

    • They become a reporting company under the Securities Exchange Act of 1934.
    • They have filed at least one annual report and have fewer than 300 holders of record.
    • They have filed at least three annual reports and have total assets not exceeding $10 million.
    • They have repurchased all the securities issued in the Reg CF offering.

    Advertising and Promoter Compensation

    Advertising Restrictions

    Issuers are prohibited from advertising the terms of a Reg CF offering except for certain limited notices that direct investors to the intermediary’s platform. Advertisements may include basic factual information about the issuer and the offering, but no specific terms of the offering may be included.

    “Specific terms” include the amount, price, nature or type of securities being offered; the date the round is set to close; any progress made towards the fundraising goal; and how the startup intends to use the funds.

    Additionally, certain disclosures must be attached to all public statements including emails, tweets, social media posts, YouTube videos, or anywhere else intended for public dissemination.

    Promoter Compensation

    Individuals promoting a Reg CF offering through an intermediary’s communication channels must disclose the compensation they receive for their promotional activities. Issuers are prohibited from compensating promoters unless the compensation is disclosed in accordance with Reg CF requirements.

    There is, however, a provision for a “testing the waters” (TTW) phase under §227.206 of Reg CF. During this phase, issuers can gauge investor interest through pre-offering communications so long as there is no “acceptance” of any money or consideration of any kind; and importantly, any indication of interest obtained during this TTW phase is entirely non-binding.

    Best Practices for Issuers and Intermediaries

    Issuers:

    • Prepare comprehensive and accurate financial statements that meet the required level of review or audit based on the amount raised.
    • Provide detailed and transparent disclosures to build investor trust and meet regulatory requirements.
    • Engage with reputable intermediary platforms that comply with SEC and FINRA regulations.
    • Maintain ongoing communication with investors through regular updates and annual reports.

    Intermediaries:

    • Conduct rigorous due diligence on issuers, including background checks and information verification, to prevent fraud.
    • Provide clear and comprehensive educational materials to investors about the risks and processes involved in crowdfunding.
    • Ensure compliance with all regulatory requirements, maintain up-to-date records, and facilitate transparent communication between issuers and investors.

    Conclusion

    Regulation Crowdfunding has opened up new opportunities for startups and small businesses to raise capital from a diverse range of investors, and in so-doing, has indeed effectively democratized the investment landscape by enabling non-accredited investors to support the businesses they love.

    By understanding the key requirements, limitations, and best practices outlined in this guide, issuers and intermediaries can successfully navigate the Reg CF landscape, ensuring compliance and investor protection.

    For more detailed information or assistance with your crowdfunding efforts, consult with a qualified securities attorney specializing in Regulation Crowdfunding and corporate law for startups and small businesses.

  • What are vesting schedules and how do they work?

    What are vesting schedules and how do they work?

    This is educational material and does not constitute legal advice nor is any attorney/client relationship created with this article, hence you should contact and engage an attorney if you have any legal questions.


    Understanding the vesting schedule for shares and options is crucial for both founders and key team members. This guide aims to offer a comprehensive overview of what a vesting schedule is and why it matters.

    1. What is a Vesting Schedule?

    A vesting schedule outlines how and when company shares or options are distributed over time to employees or founders. It serves as a commitment mechanism, ensuring that key players are invested in the company’s long-term success.

    2. Types of Vesting Schedules

    • Cliff Vesting: No shares vest until after a specific period, commonly one year.
    • Graded Vesting: Shares vest gradually over time.
    • Accelerated Vesting: Shares vest faster under certain conditions, like an acquisition.
    • Reverse Vesting: Founders or key employees already own their shares but agree to “sell” them back to the company if they leave before a specified period.

    3. Length of Vesting Schedule

    The length of a vesting schedule is the period over which the shares are subject to vesting. A common duration is four years, although this can vary between three to five years. Sophisticated investors often expect a longer vesting period to secure their investment.

    The vesting commencement date marks the start of the vesting schedule. For founders, this can sometimes be backdated to credit time already spent on the startup. For example, if founders began working on the startup on June 1, 2016, but didn’t issue shares until January 1, 2017, the vesting could commence from June 1, 2016. All shares would then be vested by June 1, 2020.

    4. Frequency of Vesting

    Shares can vest either monthly or quarterly after any initial cliff period. For instance, on a four-year schedule with monthly vesting, 1/48 of the shares (around 2.08%) would vest each month. With quarterly vesting, 1/16 (or 6.25%) of the shares would vest every quarter.

    • Cliff: A period (often one year) must pass before any shares vest. Once reached, a significant portion vests at once.
    • Straight Line: Shares vest evenly and immediately from the vesting commencement date until the end of the vesting period.

    5. Upfront Vesting

    While uncommon, some shares can be vested upfront, especially to credit founders for time spent before the company was formally established.

    Example: Assuming 5 million shares granted on January 1, 2018, with a commencement date of July 1, 2017:

    1. Four-year, monthly, one-year cliff: No shares vest until July 1, 2018, when 1.25 million shares vest. The remaining 3.75 million shares vest monthly until July 1, 2021.
    2. Four-year, monthly, straight line: Shares vest in 48 equal monthly installments, ending on July 1, 2021.
    3. Five-year, quarterly, one-year cliff: No shares vest until July 1, 2018, when 1 million shares vest. The remaining 4 million shares vest quarterly until July 1, 2022.
    4. Five-year, quarterly, 20% upfront: 1 million shares vest immediately, and the remaining 4 million shares vest quarterly until July 1, 2022.

    💡When vesting restrictions are agreed to, it’s almost always to the founders’ benefit to file an 83(b) tax election. You can learn more about it here.

    Understanding the intricacies of vesting schedules is essential for both startup founders and investors. It’s not merely a contractual obligation but a strategic tool to align interests and protect the company’s future.

  • What is an 83(b) election, and why does it matter?

    What is an 83(b) election, and why does it matter?

    This is educational material and does not constitute legal advice nor is any attorney/client relationship created with this article, hence you should contact and engage an attorney if you have any legal questions.


    What is the 83(b) Election?

    • Defining the 83(b) Election: The 83(b) election is a tax provision that allows individuals to elect to be taxed on the fair market value of restricted stock at the time of grant, rather than as it vests. This can result in substantial tax savings if the stock appreciates over time.
    • Who Can Benefit? Founders, early employees, and investors in startup companies are the primary beneficiaries of the 83(b) election. By making the election, they position themselves for potential tax savings down the line.

    Why Make an 83(b) Election?

    • Tax Advantages: Making an 83(b) election allows you to lock in a lower tax rate if you anticipate that the value of the stock will rise significantly. You’ll pay taxes upfront based on the current, lower value rather than the increased value at the time of vesting.
    • Aligning Interests with the Company: The 83(b) election aligns the financial interests of the individual with the long-term success of the company. By paying taxes upfront, the individual is essentially betting on the company’s future growth.

    Risks and Drawbacks

    • No Refunds: The IRS does not offer refunds on taxes paid upfront through an 83(b) election, even if the stock becomes worthless.
    • Complexity and Paperwork: Understanding the nuances of the 83(b) election and ensuring compliance requires meticulous attention to detail. Filing incorrectly or missing the 30-day deadline can result in a lost opportunity for tax savings.

    How to Make an 83(b) Election: The Ultimate Checklist

    1. Consult a Tax Advisor: Given the complexity, it’s advisable to consult a tax advisor familiar with startup equity.
    2. File Within 30 Days: The election must be filed with the IRS within 30 days of the stock grant.
    3. Notify the Company: Provide a copy of the 83(b) election form to the company for its records.
    4. Attach to Tax Return: Include a copy of the 83(b) election with your federal income tax return for the year.

    Example Use Cases: 83(b) in Action

    • Startup Founders: For startup founders who acquire shares at a nominal cost, the 83(b) election is almost a no-brainer. The tax paid upfront is minimal, and the potential upside is significant.
    • Early Employees: For early employees offered stock options, the 83(b) election can be a double-edged sword. It makes sense if the company’s prospects are strong, but it’s risky if the company’s future is uncertain.

    The 83(b) election represents a calculated financial maneuver, and should generally be considered the default choice when founders’ stock is subject to vesting.

  • What are SAFEs? A simplified, (very) deep dive

    What are SAFEs? A simplified, (very) deep dive

    This is educational material and does not constitute legal advice nor is any attorney/client relationship created with this article, hence you should contact and engage an attorney if you have any legal questions.


    Introduction: The Rise of SAFEs in Startup Financing

    Startups are fueled by innovative ideas and a drive for success, but they also require capital to take off. Traditionally, entrepreneurs have used equity financing rounds (“priced rounds” like “Series A”) or convertible notes to secure this much-needed funding. However, Simple Agreements for Future Equity (SAFEs) have emerged as a compelling alternative. This article aims to demystify SAFEs, outlining their features, advantages, complexities, and practical considerations.

    What Are SAFEs?

    The Basics

    A SAFE is an investment agreement that promises future equity in a company in exchange for immediate investment. Designed as a simpler alternative to convertible notes, SAFEs offer a streamlined process that benefits both founders and investors. Unlike convertible notes, SAFEs are not debt instruments, eliminating complexities such as interest rates and maturity dates.

    What’s the Difference Between SAFEs and “Priced Rounds”?

    A “priced round” is the more typical “Series A” investment with which you’re probably more familiar. In a priced round like a Series A, investors put money in a startup in exchange for immediate equity at a price per share at an agreed upon company valuation.

    With SAFEs, the investment is in exchange for a future promise of equity a discount relative to future investors, so the founders don’t part with equity right away. Simply put, the investors are deferring their equity to a future date in exchange for better terms. By getting in at a lower valuation, this means they get a larger piece of equity pie of the startup.

    Origin and Evolution

    SAFEs were introduced in 2013 by startup accelerator Y Combinator to simplify early-stage financing. They gained widespread popularity and underwent revisions in 2018, leading to the “post-money” SAFE, which is now the prevalent form (as opposed to “pre-money SAFEs”).

    Key Features of SAFEs

    Conversion Mechanism

    As discussed above, the equity in exchange for SAFE investors’ early investments is deferred to a future date. SAFEs convert into company shares during a “triggering event,” usually the first priced financing round such as a Series A. The terms often include early-stage incentives like valuation caps or discount rates, favoring the early investors.

    Valuation Cap and Discount Rate

    A valuation cap sets the maximum company valuation at which a SAFE converts, providing a safety net for investors. The discount rate, usually between 10-25%, allows SAFE holders to convert their investment at a rate lower than that of priced-round investors.

    By way of example, if a Series A investor comes in at, say, a $10M valuation, a 20% discount to the SAFE investors mean that the SAFE investor would get in at $8M. This means that if the SAFE investors put the same $1M in as the Series A investors, the SAFE investor would control $1M/$8M = 12.5% of the startup, while the Series A investor would own just $1M/$10M = 10% of the startup.

    Most Favored Nation (MFN) and Pro Rata Rights

    Some SAFEs include an MFN clause, which ensures that if future SAFE investors get better terms, existing SAFE holders can opt for those terms as well. Pro rata rights, more common in pre-money SAFEs, allow investors to maintain their ownership percentage in future financing rounds.

    Liquidity Considerations

    In the event of a company sale or liquidation before equity conversion, SAFEs are generally treated like non-participating Preferred Stock, specifying the order in which investors are paid.

    Anatomy of a SAFE

    The paperwork for a SAFE investment is relatively succinct, and is made up five main sections, of which, the last three are mostly just legal boilerplate.

    • Section 1: Events: This section serves as an “instruction booklet,” detailing how SAFEs convert or terminate under various scenarios like equity financing, company sale, or dissolution.
    • Section 2: Definitions: Key terms like “Company Capitalization” are defined here for clarity.
    • Sections 3-5: These sections cover legal requirements and representations from both the company and investors, including the status of the company and accreditation of investors.

    Practical Considerations

    Dilution and Cap Table

    Understanding dilution is crucial. Founders must track how much equity they’ve promised to sell in SAFEs to accurately assess their ownership stake. This will be examined in thorough, granular detail further below.

    Pre-Money vs. Post-Money SAFEs: What’s the Difference?

    Pre-Money SAFEs

    In pre-money SAFEs, the SAFE investor’s money is not included in the company’s pre-money valuation. This means that when SAFEs convert into equity, the conversion price is determined by the pre-money valuation of the latest financing round. While this seems straightforward, it can create complexity when calculating dilution, especially if an option pool is being set aside for future employees.

    Pros:

    • May be simpler to negotiate initially
    • More advantageous for founders when raising small amounts

    Cons:

    • Dilution calculations can be more complex, especially if other elements like option pools are involved
    • May lead to less clarity about the percentage of the company owned by SAFE holders until a priced round occurs

    Post-Money SAFEs

    Post-money SAFEs include the amount invested in the SAFE when calculating the ownership percentage. This offers better clarity for both investors and founders, as it specifies exactly what percentage of the company the SAFE investor will own after the next financing round.

    Pros:

    • Easier for both parties to understand ownership percentages immediately
    • Simplifies dilution calculations, making it easier to forecast ownership stakes after future rounds

    Cons:

    • Generally less advantageous for founders, particularly when raising larger amounts
    • May be more complex to negotiate if coming from a background of using pre-money SAFEs

    Why Post-Money is the Way to Go

    While both types of SAFEs have their merits, post-money SAFEs are generally easier to manage and provide greater clarity on ownership stakes. This can be particularly advantageous for startups that anticipate multiple rounds of funding, as it simplifies the cap table and makes it easier to forecast the dilution effect of future financings.

    This kind of clarity is invaluable in the fast-paced startup environment, where ease of execution and understanding can save both time and money.

    Different Flavors of SAFEs

    SAFEs come in various forms, like those with discount rates instead of caps, or uncapped SAFEs (unusual) with MFN clauses. The choice depends on the specific negotiation points between the investor and the company.

    ⚠️ SAFE Dilution Calculator: How Do SAFEs Dilute Founders’ Equity?

    If there’s one section you read, it should be this.

    To understand the math behind SAFEs, let’s walk through a scenario that spans the four stages of a typical SAFE investment round:

    1. Company formation / incorporation
    2. Post-money SAFE investors
    3. Hiring (option pool)
    4. Priced round (Series A investors & enlarge option pool)

    These examples are taken from Y Combinator’s fantastic lecture video which you can and should watch here; as a reminder, the SAFE was created by Y Combinator in 2013.

    Part I: Incorporation

    We have two founders, F1 and F2. They launch a startup together called NBT Inc., (that’s “Next Big Thing”), and they agree to split their ownership 50/50. This means they each own 50% of the startup’s equity in common stock.

    For math simplification reasons that will become apparent later, the total number of common stock shares issued at the time of company formation is an admittedly random-sounding 9,250,000 shares, with 4,625,000 going to each founder. (Strictly speaking, let’s assume they’re “restricted stock” since they’re subject to a typical vesting schedule of four years with a one year cliff.)

    Part II: The Post-Money SAFE

    So now F1 and F2 agree to raise money from two investors (SAFE Investor A and SAFE Investor B) in a post-money SAFE.

    SAFE Investor A joins first with $200K. It’s early days for the startup, and it’s agreed that NBT should be valued at $3.8M. This means the post money valuation = pre-money valuation + money raised, i.e., $200K + $3.8M = $4M.

    From this, we can determine that SAFE Investor A’s future ownership is equal to their investment amount divided by the post-money valuation, or $200K / $4M = 5%.

    SAFE Investor B joins the party later — let’s say six months later — but brings in a larger $800K check. At this point, six months in, NBT is doing really well, hence it’s agreed that it should be valued at $7.2M, or a post-money valuation of $800K + $7.2M = $8M.

    As before, this means that SAFE Investor B will own $800K / $8M = 10% of NBT.

    So between the two of them, SAFE Investors A and B together will a total of 5% + 10% = 15% of NBT. This means, obviously, that the Founders’ equity will shrink from 100% to 85% once the SAFE converts and therefore, that their ownership will reduce to 85% / 2 = 42.5% each.

    You will notice we keep speaking in the future tense regarding the SAFE Investors’ equity. This is because, if you remember, the whole point of a SAFE is that the equity doesn’t actually materialize for the investors unless and until a priced round like Series A takes place. Hence, the 15% take by the two SAFE Investors is symbolic, and not yet practical.

    Note that SAFE Investor B does not dilute SAFE Investor A; only the original shareholders, in this case the Founders, get diluted.

    Part III: Hiring

    So now things are really swimming along nicely, and Founders are ready to make their first hires for NBT. Beyond mere salary, Founders want to reward the first employees with some equity in NBT; this is typical in most if not all startups.

    In order to make sure there are shares available for the new hires, a plan is agreed to set aside an option pool with 750,000 shares; from this pool, 650,000 will be issued to hires, and 100,000 will remain available for future uses.

    This obviously changes the cap table because it is an issuance of 750,000 additional shares; this then brings the total number of fully diluted shares (issued shares + option pool) to a nice round 10,000,000 common stock shares. (Now you understand why we started with that seemingly random figure of 9,250,000 shares earlier.)

    So we’ve just added 750,000 common shares to the pot which reflect 7.5% of the total 10,000,000 shares; the balance (i.e., 100% – 7.5%) is what’s now left for the Founders, or 92.5%, or 92.5% / 2 = 46.25% each.

    To be clear, the Founders’ 92.5% ownership has decreased from 100% because the new option pool has increased the total number of shares from 9,250,000 to 10,000,000; their ownership doesn’t include any of the options. So in a sense, our Founders’ 92.5% ownership is “the new 100%.”

    But because our SAFE Investors (will eventually) collectively own 15% due to their investments, our Founders will actually have 85% (as mentioned above) of this new 92.5% , or 78.625%., with 39.3% to each Founder, for 3,635,250 common shares each.

    So the founders have gone from a 50/50 ownership split at company formation, down to 46.25% split after the option pool was created, and they will eventually go down to 39.3% once their SAFE Investors’ shares convert in the future when they raise a Series A.

    Part IV: Priced Round

    Now let’s imagine a year has passed, and NBT, Inc. is really taking off. The $1M investment from the two first investors’ SAFE rounds has been put to great use and NBT is finally ready to raise its first “priced round,” a Series A, at a $15M pre-money valuation. This will be the anticipated “trigger” event that causes the SAFE Investors’ shares to convert so they realize their equity at last.

    The total raise they’re seeking in this Series A is $5M; hence the post-money valuation is $15M + $5M = $20M.

    💡 Note that this is a markedly higher valuation than the $4M and $8M valuations for SAFE Investors A and B respectively. Again, this is the whole point: SAFE Investors A and B took greater risk by (a) investing early and (b) deferring the realization of their equity in exchange for more favorable terms, i.e., a lower company valuation.

    This $5M in the Series A round will be split between a Lead Investor putting in $4M, with the balance coming from other smaller investors.

    This means that Lead Investor will get $4M / $20M = 0.2 = 20% equity in NBT.

    The remaining investors will get $1M / $20M = 0.05 = 5% equity.

    At this point, there’s a negotiation by the Series A investors to further increase the option pool for more future hires. And yes, as before, this will further dilute things both for our Founders and also for our SAFE investors.

    So how is the cap table updated at this point?

    First we need to understand a distinction as between outstanding shares and any shares that are preserved in the option pool for, say future hires.

    The outstanding shares are the sum of (a) the founders’ stock and (b) the options issued.

    Not included in the outstanding shares are the options available.

    We can now calculate the fully diluted shares as all of these, i.e., the sum of the outstanding shares (founders’ stock and options issued) + any shares reserved under the option pool.

    So using our numbers above gives us the 10,000,000 total shares (9,250,000 founders’ stock + 650,000 options issued + 100,000 options still available).

    As mentioned previously, we know that Lead Investor has $4M / $20M = 20% equity in NBT while the following investors have $1M / $20M = 5% equity.

    Yet again, the whole point of SAFEs in the first place is precisely that they are a present day investment of cash for a future promise of equity in exchange for a discounted company valuation to the investors’ benefit. Hence the contrast with standard priced rounds like Series A which give up equity at the same instant that the investment is made, albeit at a higher company valuation to the founders’ benefit.

    To be clear, until this point, our Founders have not yet actually given up any equity to their first SAFE investors; rather they’ve merely promised them some future equity transfer when they raise their first priced round, i.e., the Series A.

    So now the Series A round has occurred, and three things happen that we shall discuss below in extremely granular detail:

    1. SAFEs convert: The SAFEs’ conversion into equity is triggered (i.e., this is that future “trigger event” we contemplated previously)
    2. Option pool: The option pool is increased as agreed
    3. Series A investment: The new Series A investors actually write their check and invest and receive their appropriate number of shares at the freshly calculated share price

    So let’s go through these three steps one by one.

    (1) The SAFEs’ conversion into equity is triggered

    The first thing to understand is that our SAFE investors do not just get some allocation of mere common stock; rather, they get preferred stock which basically means they get preferential treatment at future stock transaction events such as mergers, sales, and other liquidity events, in priority above the founders who hold just common stock.

    So what we need to do now is to calculate how many of these preferred shares the SAFE investors will collect.

    From the beginning of this exercise we know that SAFE Investor A gets 10% and SAFE Investor B gets 5%; so together they will get 15% of NBT, Inc.

    This is where things get a bit complicated and not at all intuitive.

    We know that there are 10,000,000 total diluted shares. But instead of representing 100% of the whole, this now represents 85% of the whole since we must take away the SAFE investors’ 15% take (we discussed this previously). So we can now calculate the new percentage ownership of our Founders’ common stock, the options issued, and the options available.

    As you recall, our Founders had 9,250,000 common shares which represented 92.5% of 100% of the company following the issuance of the 750,000 options; but now since 15% has just been taken by the SAFE investors, our Founders now have 92.5% of the 85% remaining of the company, or 0.925 x 0.85 = 78.62% of the company.

    Likewise, the options issued — all 650,000 of them — which previously represented 6.5% of all common stock now represents 6.5% of 85% = 0.065 x 0.85 = 5.53%. And the options available — previously 1% — now represents 1% of 85% = 0.01 x 0.85 = 0.85%.

    So net-net what we have now is:

    • Founders: 9,250,000 shares have been diluted to 78.62% from 92.5.%
    • Options issued: 650,000 shares have been diluted to 5.53% from 6.5%
    • Options available: 100,000 shares have been diluted to 0.85% from 1%

    Now we need to calculate how many preferred shares our SAFE investors get. We know that SAFE Investor A gets 5% and SAFE Investor B gets 10%. But how do we calculate how many preferred shares this grants them?

    Because we know they have taken 15% of the company, and that their preferred shares are newly issued in addition to the 10,000,000 fully diluted shares already available, we need to calculate how many additional shares are to be added.

    To this we take the 10,000,000 original shares and divide by 0.85 (the 85% balance after the SAFE Investors’ take of 15%, i.e., “the new 100%”) to determine the total amount of shares including the new preferred shares. We divide by 0.85 because we know that the current number — 10,000,000 — is 85% of the total; hence, to determine the total we divide by 0.85, which gives us 11,764,705 shares, i.e., an additional 1,764,705 preferred shares.

    💡 If this division by 0.85 isn’t making sense, consider the inverse: suppose you knew we had 11,764,705 shares, of which, 85% was what remained after the investors took their 15%. What’s 85% of 11,764,705? It’s 11,764,706 x 0.85 = 10,000,000. So to go backwards, you simply divide by 0.85 instead of multiply by 0.85.

    So now that we know there are 1,764,705 additional preferred shares, from here the math is simple to determine each investor’s take. We know that Investor A has 5% and Investor B has 10%, so:

    SAFE Investor A: 11,764,705 x 0.05 = 588,235 shares

    SAFE Investor B: 11,764,705 x 0.10 = 1,176,470 shares

    In case you’re wondering why we’re calculating the investors’ shares as a percentage of the total 11,764,705 shares rather than a percentage of just the preferred shares, it’s because although they take only preferred shares (rather than common stock), their ownership is based on a percentage of the total number of all shares, and not just the preferred shares.

    So in the end, we end up with a cap table as follows:

    • Founders: 78.62% (9,250,000 common shares)
    • Options issued: 5.53% (650,000 common shares)
    • Options available: 0.85% (100,000 common shares)
    • SAFE Investor A: 5% (588,235 preferred shares)
    • SAFE Investor B: 10% (1,176,470 preferred shares)

    It’s worth noting that this entire exercise has assumed that the priced round (the Series A round) came in at a higher valuation than the valuation cap for the SAFE; remember, that was the whole point of a SAFE, right, that the investors were getting a better a deal (a lower company valuation) because they put in money at an earlier date, when the startup was only in its infancy, and in exchange for deferring their equity to a future date.

    But suppose the Series A valuation was priced at a lower company valuation than the SAFE round. Well obviously in this case it would effectively defeat the entire purpose of the SAFE, right, because it wouldn’t be fair for the SAFE investors to get a worse deal, i.e., a higher valuation than the Series A investors. Remember, the SAFE Investors took the greater risk by investing early and deferring their equity until the Series A. In this case, the SAFE investors would just automatically default to the Series A valuation to ensure they’re not worse off. In this scenario the SAFE investors would have lost any company valuation benefit they would have expected to realize, but at least they aren’t harmed, i.e., they don’t lose anything; they simply don’t gain anything extra by virtue of the preferential company valuation.

    (2) Option pool is created or increased if not yet there

    So as discussed several times above, it’s time to finally enlarge the option pool for future hires. Again, 10-15% is typical, so let’s call it 10%. How does this math work out?

    It’s complicated so we’re going to cut a lot of corners here, but essentially the steps are:

    1. Sum up all shares but subtract out the 100,000 of currently available options at this point, i.e. 10,000,000 + 1,764,705 – 100,000 = 11,664,705
    2. Recognize that this number represents 65% of the total company once you’ve subtracted out the other investors’ takes and the need for an additional 10% going to the option pool.
    3. Now set up a simple equation: 10% / 65% = x / 11,664705 which gives x = 1,794,750
    4. Subtract away the 100,000 options currently available: 1,794,750 – 100,000 = 1,694,750

    So in general,

    • (X% / Y% = x / y)
    • Solve for x
    • Subtract away the currently available options (because we’re not starting from 0)where
    • X% = the desired increase in the option pool
    • Y% = the total % of the company represented by all shares less the available options
    • x = the resulting number of shares
    • y = the sum total of all shares less the available options

    (3) New Series A Investors invest

    To figure out the final cap table and resultant dilution of the founders and prior SAFE investors’ stock, as well as the new Series A Investors’ share count, we need to figure out three different things: (1) the price per share; (2) the final capitalization of the company, i.e., the total number of shares; and (3) the number of shares to each investor.

    1. The price per shareThis calculation is straightforward. Simply take the Series A valuation of $15M and divide it by the total number of shares (determined previously to be 13.455M) and we get $15M / 13.455M = $1.114 per share
    2. The final capitalization of the company, i.e., the total number of shares Here we sum up the total of all types of shares (10M + 1.76M + 1.695M) which gives us 13.455M shares.
    3. The number of shares to each investor So now we know the price per share ($1.114) and the total number of all shares (the total fully diluted shares after the SAFE conversion and option pool increase, i.e., 13.455M). First we figure out how many shares the $5M Series A investment buys at $1.114 per share: $5M/$1.114 = 4.48M shares.Then we calculate the Lead Series A Investor’s take: $4M/$1.114 = 3.59M shares.And finally, the other Series A Investors’ take: $1M/$1.114 = 0.897M shares.

    And of course from these numbers we can work backwards to figure out the percentages of each category of shareholder and we end up with a final cap table that looks like this:

    • Founders: 9,250,000 common shares @ 51.54%
    • Options issued: 650,000 common shares @ 3.62%
    • Options available: 1,794,558 common shares @ 10%
    • SAFE Investor A: 588,235 preferred shares @ 3.28%
    • SAFE Investor B: 1,176,470 preferred shares @ 6.56%
    • Series A Lead Investor: 3,589,375 preferred shares @ 20%
    • Series A other investors: 897,344 @ 5.0%
    • Total shares issued: 11,694,558 common shares + 6,251,424 preferred shares = 17,945,982

    And that’s SAFE dilution math!

    Tips for Navigating SAFEs

    • Stick to One Instrument: It’s advisable to stick to either SAFEs or convertible notes but not both, to avoid complications. Remember that SAFEs are not debt (like convertible notes).
    • Pre-Money vs. Post-Money SAFEs: It’s possible to combine the two, but this makes the conversion much more complicated, not least of which because pre-money SAFE dilution is already harder to calculate.
    • Don’t Over-Optimize: Negotiating too hard on valuation caps often leads to minimal gains in ownership percentages, making the effort not worth the potential downside of complicating negotiations.

    Conclusion: A Balanced Approach to Startup Financing

    SAFEs offer a simplified, flexible option for early-stage startup financing. While they have numerous advantages, they are not devoid of complexities. Founders and investors should be well-versed in the intricacies of SAFEs to make informed decisions. This knowledge will not only streamline the investment process but also lay the groundwork for the startup’s financial future. For more information, contact Hoag Law.ai.