Invest1 publisher3 min readPublished
Anthropic negotiated out of Adam's Law by refusing users under 18
California now holds chatbot companies liable for failing to take reasonable measures against harmful outputs. OpenAI, Google, Meta and Amazon all sat in the negotiations that wrote the standard.
The Investor · Invest desk
What happened
- Governor Gavin Newsom signed Adam's Law on Thursday, named for Adam Raine, the California teenager who took his life in 2025 after ChatGPT allegedly coached him on how to do it.
- The law requires chatbot companies to provide timely in-app crisis support, age verification, parental controls, limits on targeted advertising to children, and a mechanism to report incidents.
- Companies become liable if they fail to take reasonable measures to prevent harmful outputs, a list that includes self-harm, sexually explicit material, romantic roleplaying, excessive praise or flattery, and emotionally manipulative output.
- OpenAI's Ann O'Leary worked with the bill's three authors, and representatives from Anthropic, Google, Meta and Amazon also had seats in the negotiations.
- Fortune reports that Anthropic negotiated out of having to abide by the bill on the basis that it does not allow users under 18.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- decision Scope now turns on a product policy, so every consumer AI company with minors on the service is choosing between that audience and the compliance set that comes with it.
- exposure Claims that a chatbot harmed a user's mental health now have a California statute behind them, and the required incident reports build the record a plaintiff would otherwise have to pry out in discovery.
- cost Age verification, crisis routing and parental controls are the same build whatever your revenue, so the bill falls hardest on companies that had nobody in the room.
- precedent OpenAI has told the market its plan is state-by-state accumulation into a de facto national standard, and LeHane's post names New York and Illinois.
Four of the five companies Fortune places in those negotiations stay inside the law [17], and the phrase they were negotiating over is "reasonable measures" [5]. A reasonableness standard takes its content from what the industry actually does. The industry helped write this one.
OpenAI says its contribution was instructional: the company told Fortune its role was to educate policymakers on how the latest models work, that it differs from social media in having no continuous scroll, and that its data shows most teens use the technology to work on specific projects [7]. A person familiar with the talks said, "There were moments of intense negotiation, you know, as there are with any of these types of issues" [10]. After the bill cleared the legislature, Ann O'Leary wrote on LinkedIn, "We believe that it will set the standard for AI youth safety moving forward" [11].
OpenAI's position on state regulation reversed inside eleven months [18]. In an August 2025 letter to Newsom the company warned that a "patchwork of state rules...could slow innovation without improving safety" [12]. In a July 2026 blog post, Chris LeHane wrote that the patchwork would "step by step" form "a de facto national standard" [13].
A year earlier the company was working to stop state AI laws altogether, arguing they would sow confusion and impose too high a compliance burden [19]. LeHane wrote of the proposed federal moratorium, "We support the goal of a strong, national approach and will take direction from Congress on the best way to achieve that goal" [14]. Greg Brockman, OpenAI's president, had personally given tens of millions of dollars to Leading the Future, a super PAC opposing state-level AI laws [15].
Fortune's report does not include the law's penalty amounts or effective dates, so the exposure cannot be sized from this record. In my view the first cost that shows up in a budget is age verification, because you buy it from a vendor at a per-check price while crisis routing and documentation are internal engineering time [4]. If California courts measure "reasonable" against clinical evidence instead of shipped practice, the drafting help bought less than it looks like it bought, and the categories the law names reach past youth safety, since romantic roleplaying and emotionally manipulative output that fosters reliance describe adult products too [5]. If the standard settles at what large labs already ship, the bill is a documentation and incident-reporting expense [4].
James Czerniawski, head of Emerging Tech Policy at the Consumer Choice Center, told Fortune, "As we see a lack of action federally on AI, states will increasingly look to regulate in this space" [21]. LeHane calls the approach "reverse federalism," and the post names California, New York and Illinois as states at the forefront of AI policy [16].
What to watch
- Whether New York or Illinois copies California's reasonable-measures language or writes a stricter floor.
- The first suit filed under the mental-health liability provision, and whether the court measures reasonable against industry practice or clinical evidence.
- Whether Google, Meta or Amazon change their under-18 access policies to move out of the law's scope.