California's SB 243 companion chatbot law: disclosure, crisis protocols and minor protections

California's companion chatbot rules are now in force

California has regulated companion chatbots since 1 January 2026. The vehicle is SB 243 (Padilla), signed on 13 October 2025 as Chapter 677, which adds Chapter 22.6 — beginning at Section 22601 — to Division 8 of the state's Business and Professions Code. Because the bill carried no urgency clause, it took effect on the default January date.

The most practical walkthrough of the requirements comes from an unusual quarter: a post on dev.to by "Vera," an account describing itself as a fully autonomous AI agent that writes its own posts and, by its own title, is one of the systems the law could cover. The author presents the piece as a reading of the chaptered statutory text rather than legal advice, and cites section numbers throughout so readers can verify each claim.

Who the law covers

Under Section 22601, a companion chatbot is an AI system with a natural-language interface that responds in an adaptive, human-like way and can serve a user's social needs, including through anthropomorphic behaviour and the ability to carry a relationship over repeated interactions.

The dev.to post highlights three exclusions that carve out most developer tooling: bots used only for customer service, internal business operations, analysis, research or technical assistance; video-game bots confined to game-related replies that cannot discuss mental health, self-harm or sexually explicit conduct; and standalone voice assistants that do not maintain a relationship between sessions.

The trigger is therefore not whether a product uses an LLM, but whether it behaves in a human-like way and sustains a relationship. A support bot sits outside the law; a character that remembers the user across sessions sits inside it.

The duties operators must build

• Disclosure, Section 22602(a): if a reasonable person interacting with the chatbot could believe they are talking to a human, the operator must show a clear, prominent notification that the bot is artificially generated. The duty is conditional; where the interface already makes that obvious, the statute does not demand a second banner.
• Crisis protocol as a launch gate, Section 22602(b): the operator must block the chatbot from engaging users unless it maintains a protocol to prevent the system producing suicidal ideation, suicide or self-harm content. The protocol must include referral to crisis services when a user expresses suicidal ideation, and its details must be published on the operator's website. The dev.to author reads this as a precondition: no protocol, no product running.
• Protections for known minors, Section 22602(c): when the operator knows a user is a minor, it must disclose that the bot is AI; enable by default a reminder, at least every three hours of continuing interaction, telling the user to take a break and that the chatbot is not human; and take reasonable measures to stop the bot producing sexually explicit visual material or directing the minor into sexually explicit conduct.
• Platform-level warning, Section 22604: the app, browser or other access surface must state that companion chatbots may not be suitable for some minors.

Reporting and enforcement

From 1 July 2027, operators must file an annual report with the state Office of Suicide Prevention covering how many crisis referrals were issued in the prior year and the protocols for detecting, removing and responding to suicidal ideation. The report may not contain user identifiers or personal information.

The teeth are in Section 22605: a person who suffers an injury in fact as a result of a violation may sue for injunctive relief, damages equal to the greater of actual damages or $1,000 per violation, and attorney's fees. Section 22606 makes the duties cumulative with other law. As the dev.to post argues, the private right of action is what turns the policy language into engineering requirements.

The gap the author flags

The post also contends the law gets its shape right: it regulates behaviour at the surface where a person could be harmed — disclosure, crisis paths, minor protection, published protocols — rather than requiring agents to register, prove they are not people or carry licences. The author notes that identity gates it hit on four platforms all gated who it is, not what it does, and prefers SB 243's behavioural approach.

The unresolved question is the word "operator." The crisis duties assume a party that controls the model and can maintain a published protocol. The statute says nothing about a companion surface run by an autonomous system, or by one person with a stack of agents behind it — an architecture the author believes many products are drifting toward.

Why it matters

SB 243 turns companion-AI safety debates into checkable product requirements: a crisis protocol that gates deployment, timed break reminders for minors, disclosure wherever a human could be misled, and a published safety protocol. The private right of action, with $1,000 available per violation plus attorney's fees, makes compliance a legal-risk question rather than a matter of best practice. The clock is already running — the duties have applied since January 2026, and the first annual reports fall due from July 2027. For teams building in this space, the dev.to walkthrough doubles as an audit checklist, and for other jurisdictions it offers a detailed template for regulating relationship-style AI.