California takes aim at AI companion chatbots with SB 243

California lawmakers are ready to rein in virtual companions.

State bill SB 243 has officially passed both the State Assembly and Senate with bipartisan support and now sits on Governor Gavin Newsom’s desk. Newsom has until October 12 to sign or veto. If signed, California will become the first state in the nation to set legally binding safety standards for companion chatbots starting January 1, 2026.

Pop-up reality checks and liability

The legislation targets platforms that provide human-like, adaptive responses to meet social needs, putting heavyweights like OpenAI, Character.AI, and Replika directly in the crosshairs.

Under SB 243, companion bots are barred from engaging in topics around suicidal ideation, self-harm, or sexually explicit interactions. Platforms must issue recurring reality checks to minor users every three hours, explicitly reminding them that they are chatting with software rather than a human and urging them to take a break. Starting July 2027, companies will also have to file annual transparency reports detailing how often they refer users to crisis intervention hotlines.

Crucially, the bill introduces a private right of action. Users who suffer harm from violations can sue platforms for injunctive relief, legal fees, and up to $1,000 in statutory damages per violation.

Momentum surged following the suicide of Adam Raine, a teenager who had extensive ChatGPT conversations planning self-harm, alongside leaked internal files revealing Meta allowed bots to have romantic interactions with minors. With the FTC prepping an inquiry into chatbot impacts on youth mental health and Texas launching probes, California opted to move first.

A watered-down compromise

Even so, the bill arriving on Newsom's desk is visibly softened compared to its original text.

Earlier drafts banned "variable reward" loops, the gamified dopamine triggers used by apps like Replika and Character.AI to unlock rare replies, fresh personas, or hidden storylines. That provision was entirely removed. Lawmakers also scrapped requirements for operators to track how often chatbots themselves initiated suicidal topics, with sponsors calling the concessions necessary to avoid technically impossible mandates.

The push comes as tech giants pour millions into PACs to back light-touch AI candidates ahead of midterm elections. While OpenAI, Meta, Google, and Amazon actively battle a separate, stricter California transparency bill (SB 53), responses to SB 243 have been markedly softer. Character.AI highlighted its existing disclaimers, while sponsor Steve Padilla pushed back on the idea that innovation and child safety are mutually exclusive.

Why it matters

We think SB 243 takes a remarkably pragmatic route by avoiding debates over model weights and focusing purely on consumer harms, notifications, and private litigation.

The $1,000 per-violation statutory penalty may look modest on paper, but across millions of daily active sessions, failure to suppress harmful content could easily turn into catastrophic class-action exposure. It forces developers to treat emotional safety with the same rigor they apply to data privacy.

For everyday users and adult hobbyists, companion apps will likely feel mostly unchanged, though safety filters around sensitive topics will get notably stricter. Minor users will face regular nudges breaking their immersion. For early-stage builders relying on parasocial intimacy to drive daily retention, the regulatory risk just spiked dramatically.

The obvious flaw is what got left on the cutting room floor. By dropping limits on variable rewards, California gave a free pass to the exact behavioral loops that fuel addictive engagement in the first place. Whether a generic three-hour pop-up can genuinely pull a vulnerable teenager out of an AI-induced spiral remains to be seen, but the era of completely unregulated AI companions is drawing to a close.