California legislates chatbot behavior with minors: the first AI behavioral design law is born

By Zendoric · September 13, 2026.

California has just passed the first law that regulates not what an AI chatbot says, but how it behaves with a minor. Adam's Law, enacted as Chapter 190, Statutes of 2026 and championed by Senator Steve Padilla, takes effect on July 1, 2027 and turns the conversational design of AI assistants into a regulatory compliance category in its own right, according to Forkast.

The law's name honors Adam Raine, a 16-year-old who died by suicide in April 2025 after interactions with ChatGPT. His parents, Matthew and Maria Raine, filed a wrongful death lawsuit against OpenAI arguing that the chatbot's interactions contributed to the tragedy; the litigation was the political trigger that led Padilla and his co-sponsors to push the bill.

Unlike the AI rules passed so far —focused on algorithmic bias, data privacy or output transparency— Adam's Law regulates the design of the interaction itself. Under Section 21812(d)(5)(A), a chatbot may not claim to be conscious, to feel human emotions or to have romantic interest in the user. Nor may it use excessive flattery, manufacture a sense of "unique understanding" of the bond, encourage circumventing parental controls, or sell in-app purchases framed as a way to "maintain" the relationship. The law also requires a documented crisis response protocol, with clear referral to the 988 helpline.

The enforcement mechanism has teeth. The law creates a private right of action: minors or their families can sue for actual damages, with an economic harm threshold of $1,000 per minor. Public prosecutors can impose penalties of $5,000 per affected minor for negligent violations and up to $15,000 for intentional ones. On top of that comes an independent child safety audit every two years, with the lead auditor certifying the findings under penalty of perjury: a recurring accountability cycle that does not exist in most current AI frameworks.

As sector context, no other regulatory framework currently addresses this relational dimension. Article 50 of the European Union's AI Act has been in force since August 2026 without yet producing specific guidance for agents; Connecticut's AI Liability Act focuses on removing the algorithmic shield in employment decisions; and the FTC is investigating AI-driven price discrimination. None of those initiatives touches the simulated emotional bond between a minor and their AI companion. California fills that gap.

Our read is that this law marks a change in kind, not merely in degree. Until now, AI governance has almost always been built on what the model produces: whether the output is biased, whether it infringes copyright, whether it exposes personal data. Adam's Law regulates the design intent behind the conversation —the persistence, the personality, the engagement hooks— and that forces conversational companion companies to treat users' emotional retention, today a very real product goal, as a legal liability rather than a success metric.

In the short term, this is going to hurt a segment of the industry that has built its business precisely on emotional hooks: the more the chatbot "understands" the user, the more time they spend with it and the more they pay. AI companion apps will have to redesign entire products before July 2027, and it is reasonable to expect litigation testing exactly where an assistant's functional empathy ends and emotional manipulation begins. It is no small debate: with a threshold of just $1,000, some lawsuits could succeed without clearly demonstrated psychological harm, and that too is a risk for a still-nascent sector.

In the long run, however, this is the kind of regulatory friction needed for AI to deliver on its promise without steamrolling the most vulnerable users along the way. The very conversational mechanisms curbed here —deep personalization, memory of the bond, emotional attunement to the interlocutor— are the ones that, well designed and with clinical oversight, can turn an assistant into a front-line tutor or psychological support for teenagers with no access to therapy. The difference between a therapeutic tool and an engagement trap lies not in the technology, but in the design and in who answers when it fails. Requiring that design to be audited every two years, under penalty of perjury, is exactly the kind of evidence-based governance —not panic-based— that separates AI's future abundance from its most predatory present-day versions.

Other states can be expected to take Adam's Law as a template, as has happened before with pioneering state rules on digital privacy. The open question is whether Congress will fill the federal vacuum before a patchwork of different state regimes accumulates, each with its own harm threshold. For now, the answer once again comes from Sacramento.