AI Regulation Tracker  /  State law

California's SB 243 companion-chatbot law is in force, and users can sue for $1,000 per violation

Since January 1, 2026, operators of AI companion chatbots must tell users they are not human, run suicide and self-harm protocols, and protect minors. A user injured by a violation can sue for the greater of actual damages or $1,000 per violation, plus attorney's fees.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published July 7, 2026  ·  Last updated July 7, 2026

California's companion-chatbot law is now operative. Signed by Governor Gavin Newsom on October 13, 2025, SB 243 took effect January 1, 2026, and it regulates AI companion chatbots in California as a distinct category, an approach that legal commentators have described as among the earliest of its kind in the United States. For any company that ships an AI persona designed to hold a conversation and keep a relationship going across sessions, the compliance clock has already started, and the law gives injured users a direct way to sue.

What the statute actually covers

SB 243 turns on a specific definition. Under Section 22601, a "companion chatbot" is an artificial intelligence system with a natural language interface that provides adaptive, human-like responses and is capable of meeting a user's social needs, including by exhibiting anthropomorphic features and sustaining a relationship across multiple interactions. That definition is deliberately narrow at the edges. The statute excludes customer-service bots used for a business transaction, bots contained within a video game, and standalone voice-activated virtual assistants of the kind built into consumer hardware.

The practical test for a product team is not the marketing label but the function. If your chatbot is built to be a friend, a companion, a romantic persona, or an ongoing confidant, it is likely inside the definition. If it exists to answer support tickets or run a game world, it is likely outside. Products that sit in between, such as a wellness companion or a coaching persona, need a documented analysis rather than an assumption.

The three operating duties

Section 22602 sets out the core obligations. First, when a reasonable person interacting with the chatbot would be misled into believing they are talking to a human, the operator must issue a clear and conspicuous notice that the companion chatbot is artificially generated and not human. Second, the operator must maintain a protocol for preventing the production of suicidal ideation, suicide, or self-harm content, and when a user expresses suicidal ideation or self-harm, the protocol must include a notification that refers the user to crisis service providers, such as a hotline or text line. The operator must publish details of that protocol on its website.

Third, the statute adds protections for minors. Where the operator knows a user is a minor, it must disclose that the user is interacting with artificial intelligence, provide by default a clear notice that responses are machine-generated, remind the user at least every three hours during continuing use to take a break, and take reasonable measures to prevent the chatbot from producing sexually explicit content for that minor. These are default-on requirements, not optional settings a user must find and enable.

The enforcement mechanism that changes the calculus

What gives SB 243 immediate weight is Section 22605, its private right of action. A person who suffers an injury in fact as a result of a violation may bring a civil action to recover the greater of actual damages or one thousand dollars per violation, together with injunctive relief and reasonable attorney's fees and costs. That structure does not depend on a state regulator opening a case. It lets private plaintiffs, and the plaintiffs' bar, enforce the statute directly, and the per-violation floor of $1,000 creates real aggregate exposure for a product with a large user base.

The statutory damages figure is a floor, not a cap, because the plaintiff recovers the greater of actual damages or $1,000. For operators, the combination of a per-violation measure, statutory minimum damages, and fee-shifting is the part of SB 243 most likely to drive behavior, because it converts a missing disclosure or a broken crisis referral into a quantifiable claim.

What the law does not do

SB 243 is targeted, and reading it too broadly is its own risk. It does not ban companion chatbots, and it does not require age verification of every user. It does not reach customer-service bots, in-game characters, or voice assistants that fall within the exclusions. It is a California statute, so it governs operators offering companion chatbots to users in California rather than imposing a national rule, though a company that serves a national audience will in practice apply the controls everywhere rather than gate them by state. And the annual reporting duty to the Office of Suicide Prevention, tied to the crisis-protocol provisions, does not begin until July 1, 2027, so the reporting piece is a forward obligation while the disclosure, protocol, and minor-safety duties are live now.

It is also a distinct instrument. SB 243 is not California's SB 53 on frontier-model transparency, not AB 723, and not the state's professional-conduct or civil-rights rules for AI. Teams tracking several California AI measures at once should keep the companion-chatbot duties in their own lane.

Why California's rule reaches beyond California

For a compliance lead outside California, the practical question is whether SB 243 can be ignored. In most cases it cannot. Companion-chatbot products are national or global by default, and few operators can cleanly separate California users from the rest of their base at the level each duty requires. The disclosure, the crisis referral, and the minor safeguards are cheaper to run for everyone than to gate by geography, so a California statute tends to set the operating floor for the whole product. SB 243 also arrives alongside a broader wave of state action on companion and character chatbots, so a control built for California is likely to satisfy the next state's version rather than being wasted work. Treating the statute as a de facto national baseline is the conservative reading, and it is the one most large operators are adopting.

The counterpart risk is over-application. Because the definition excludes customer-service bots, in-game characters, and standalone voice assistants, a company that runs several bot types should not apply the companion-chatbot controls indiscriminately. The right posture is a documented classification of each bot against the Section 22601 definition, so the product can show why a given assistant is in or out of scope. That record matters if a plaintiff later argues a bot was a companion chatbot in function even though it was labeled something else.

Building a record that survives a claim

Because Section 22605 counts violations, documentation is not a formality. Operators should keep evidence that each duty functions, not just that a policy exists. For the non-human disclosure, that means capturing where and when the notice appears in the user flow. For the crisis protocol, it means retaining the published protocol, the trigger logic, and proof that a referral to crisis resources is delivered when self-harm language appears. For minors, it means showing how the product determines a user is a minor and logging that the default notice, the three-hour break reminder, and the explicit-content block are active. A defensible file answers three questions for every duty: what the control is, that it was live during the relevant period, and that it actually fired. That is the difference between contesting whether a violation occurred and conceding it.

What operators should do now

The near-term work is verification rather than theory. Confirm that the non-human disclosure appears at the point a user could be confused, not buried in a settings page. Confirm that the self-harm protocol exists, is published on the website, and produces an actual referral to crisis resources when triggered. Confirm that minor users receive the default machine-generated notice, the three-hour break reminder, and the block on sexually explicit content. Because the private right of action counts violations, the difference between a control that is documented and a control that reliably fires in production is the difference between a defensible posture and a per-user claim.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.