California's companion chatbot laws, seen from inside the app: SB 243 now, Adam's Law in 2027

Two California laws now decide part of what you see inside an AI companion app. SB 243 has applied since January 1, 2026. For an adult, its visible pieces are a notice that the chatbot is not human and a referral to crisis services if you write about suicide or self-harm. SB 1119, signed on September 10, 2026 and known as Adam’s Law, is written mostly for accounts held by under-18s, and most of its duties start on July 1, 2027. For adults on an 18+ app, the part of Adam’s Law most likely to reach the screen is the age check.

This guide covers the user side: what each rule asks the app to show you, and what the 17 apps in our directory say they already do, based on their own pages as read on October 30, 2026. The full summary of both laws is on our laws page. Nothing here is legal advice, and nothing here is a finding that any app does or does not comply.

The dates that matter

Date	What happens
October 13, 2025	SB 243 signed (Chapter 677)
January 1, 2026	SB 243 duties apply
September 10, 2026	SB 1119, Adam’s Law, signed (Chapter 190)
January 1, 2027	SB 1119 takes effect; AB 1043 age signals start
July 1, 2027	Adam’s Law rules for child users start; SB 243 annual crisis-referral reports start
January 1, 2028	Attorney General’s public complaint channel due
January 1, 2029	First independent child-safety audit due (operators under $500 million in revenue are exempt before 2032)

The chaptered text of SB 243 has no start clause. California’s code record for section 22602 reads “Effective January 1, 2026,” which matches the state constitution’s rule for a bill passed before the Legislature’s mid-session recess and signed after it (Article IV, section 8(c)(2)). Hunton and Fenwick both give January 1, 2027 for SB 1119, while the bill itself says sections 21812, 21812.5 and 21813, which hold the child-user rules, become operative on July 1, 2027.

SB 1119 also rewrites SB 243’s section 22602 and drops its paragraph on users the operator knows are minors (an AI disclosure, a break reminder at least every three hours, and limits on sexual content). The bill sets no separate date for that change. If it takes effect with the rest of the act on January 1, 2027, the old minor rules end six months before the new ones start. None of the sources we read address those months.

“This is an AI”: the notice SB 243 asks for

The first rule in SB 243 is conditional. If a reasonable person talking to a companion chatbot “would be misled to believe that the person is interacting with a human,” the operator has to show “a clear and conspicuous notification” that the chatbot is artificially generated and not human (section 22602(a)). The law sets no wording or frequency for adults. A second rule, section 22604, makes every operator disclose on the app or browser that companion chatbots may not be suitable for some minors. Adam’s Law amends only section 22602, so 22604 stays as it is.

Several apps already say this in their own words. Replika’s terms, updated October 28, 2026, describe a Replika as an AI chatbot and not a human being, say it “will confirm it is an AI if you ask,” and say the service includes “periodic reminders that you are talking to an AI.” ourdream’s safety site says: “You are interacting with software, not a person.” Character.AI announced in October 2024 that it would add “a revised disclaimer on every chat to remind users that the AI is not a real person.” Janitor AI’s changelog entry of January 18, 2026 lists “AI content notice - Clear disclaimers about AI-generated content.” That entry doesn’t mention the law.

When a message mentions suicide or self-harm

This is the rule adults are most likely to run into. Under section 22602(b), a companion chatbot may not engage with users at all unless the operator keeps a protocol for preventing suicidal-ideation, suicide or self-harm content. The protocol has to include a notice that refers the user to crisis services, such as a suicide hotline or crisis text line, when the user expresses suicidal ideation or self-harm. The operator also has to publish details of the protocol on its website. Adam’s Law keeps this paragraph unchanged, and it covers every user, not only minors.

Seven apps in our directory describe what happens, and they don’t handle it the same way.

Kindroid has the most detailed public version, in section 4.3 of its Moderation Guidelines (updated October 9, 2026). Automated systems watch conversations, including the Kindroid’s own replies, and are “designed to tell real-world risk apart from venting, general discussion, and fiction.” Kindroid doesn’t act on a single detection. A second one brings up a notice listing the 988 Suicide & Crisis Lifeline (call or text 988), the Crisis Text Line (text HOME to 741741) and, outside the US, the IASP directory. If it keeps happening, normal conversation is paused, and the pause “doesn’t end on its own”; the screen tells you to email [email protected] to come back.

ourdream screens “every user message and every character reply” before delivery, according to its Wellbeing resources page (last modified September 2, 2026). If your message is flagged, the character doesn’t reply and you see a notice pointing to crisis support; “in some regions the notice is a dialog the user must acknowledge.” A flagged character reply is withheld and swapped for a filtered-content notice, and you can rewind or ask for a new reply. ourdream’s content rules also ban suicide and self-harm roleplay “including within a story,” so fiction gets no exception there.

Chai’s Safety Center says every user message is scanned in real time, a harmful reply is refused or replaced, and “when a user expresses self-harm intent, CHAI provides a notification referring the user to crisis service providers.” Replika’s terms say automated features “may recognise messages suggesting a crisis and respond by showing you crisis resources,” and name 988 for the US. Talkie’s Contact page says its content-safety mechanisms look for suicidal ideation, suicide or self-harm, and on detection the platform “presents a notification to the user and provides referrals to relevant crisis support services,” including 988 in the US. Character.AI said in October 2024 that a pop-up triggered by certain phrases directs users to the National Suicide Prevention Lifeline. Janitor AI’s January 18 changelog entry reads “Added crisis detection and safety modals.”

The law’s text says nothing about fiction or roleplay, and the apps have drawn that line in different places. Kindroid says it tries to separate fiction from real risk, while ourdream bans self-harm roleplay in fiction outright. A 3-star Janitor AI review on the US App Store, dated August 18, 2026, is headed “I can no longer do doomed stories” and blames “a new filter that censorship thing that prevents self harm or harmful role play.”

For the other apps, we found less. The terms of Nomi and Paradot tell users who feel suicidal to call 911, and Paradot’s add that they must stop using the service. For Anima, Candy.ai, CrushOn.AI, EVA AI, Grok, Nectar AI, PolyBuzz and Secret Desires, the pages we read on October 30, 2026 (each app’s terms and privacy policy, plus its FAQ, community guidelines or safety pages where it has them) did not describe a crisis-referral notice. A protocol can exist without appearing on those pages, so this is a note on what we could find on that date, not a finding about what any of these apps does.

If you are in the US and thinking about suicide or self-harm, you can call or text 988.

The age check is where adults will notice Adam’s Law

Adam’s Law gives every operator two ways to meet section 21811, including operators of adults-only apps. The first is to determine the user’s age through California’s Digital Age Assurance Act, AB 1043. The second is to give every user the protections meant for children. Under that second route, the child default settings can’t be changed unless the operator has “actual knowledge” that the user is not a child; Hunton’s summary reads the rule the same way. Those defaults are persistent memory off, push notifications off, a one-hour limit on a single session and a two-hour daily limit. An app that bars under-18s also has to publish a short description of how it handles age assurance (section 21812(b), operative July 1, 2027).

AB 1043 is where the age check moves to your phone. From January 1, 2027, operating-system providers must ask at account setup for the birth date or age of the device’s user and pass apps a signal saying whether that user is under 13, 13 to 15, 16 to 17, or 18 and over. Apps must request that signal when they are downloaded and launched. For devices set up before 2027, the operating system has to offer a way to enter a birth date or age before July 1, 2027.

Nine of the 17 apps in our directory rely on a self-declared age or don’t say how they check. The other eight describe more, according to their own pages, checked October 27, 2026 (Replika’s row October 30, 2026):

App	What its pages say about checking age
Chai	Persona selfie age estimate or government photo ID; a paid subscription is required in some regions
Character.AI	Account signals first; a Persona selfie only if flagged as under 18, and ID only as a last step
Kindroid	A payment on file counts; flagged accounts subscribe or verify with Yoti (selfie passes at an estimated 21 or older, otherwise ID)
Janitor AI	k-ID check (face estimate, ID scan or AgeKey) in Australia, Brazil and the UK only
ourdream	Third-party check (Go.cam or AgeVerif) where law requires it, including the UK and Australia
PolyBuzz	Birth date for everyone; ID or face check in the UK (Ondato, Yoti)
Replika	Privacy Policy of October 28, 2026: where required by law or where a declared age is in doubt, an age-estimation or age-verification check by selfie or ID, processed on the device or by Incode
EVA AI	Self-declared; accounts the system suspects are under 18 are blocked until proof of age

The nine are Anima, Candy.ai, CrushOn.AI, Grok, Nectar AI, Nomi, Paradot, Secret Desires and Talkie. Nomi’s terms say it “intends to, as required and in a way that best respects user privacy, implement additional, age-verification measures.”

A US App Store review of Character.AI dated October 17, 2026 opens: “you have to provide ID and age verification now.” The company’s FAQ says a selfie is asked for only when it believes a user is under 18.

Apps whose own terms allow under-18 users

Three apps in our directory say outright that under-18s can use them. Grok’s terms set 13 as the minimum, with a parent’s or guardian’s permission for 13 to 17. Sources differ on whether Grok still has companions: xAI’s Grok FAQ, read on October 30, 2026, says Companions are available on the iOS app only, while the @grok account on X said on September 10, 2026 that xAI retired the feature around September 7. Talkie’s terms set 14, higher in some countries. Character.AI keeps an under-18 experience without open-ended chat. Three more are unclear. Anima’s terms say the service is not intended for under-18s but also allow minors with a parent’s permission and supervision. PolyBuzz’s terms bar under-18s while its FAQ describes a “Teen Mode.” Nectar AI’s Terms of Use (version 2.2, August 15, 2026) say users under 18 “may only access SFW content areas and features,” while its FAQ says users under 18 “are not permitted to use Nectar under any circumstances.” Whether a given app meets the law’s definition of a companion chatbot is a legal question this post doesn’t answer.

Under SB 243 today, a user the operator knows is a minor gets an AI disclosure, a default reminder at least every three hours to take a break, and reasonable measures against sexually explicit images or the chatbot urging sexual conduct. From July 1, 2027, Adam’s Law asks much more of operators that let children in. The main items, from sections 21812 and 21813:

• Defaults that only a parent can change: memory off, push notifications off, one hour per session, two hours per day, and a parent’s power to block access for a child under 16.
• An AI notice in language a child understands, repeated during long chats.
• Reasonable measures so the chatbot doesn’t express romantic interest, claim to be sentient or human, encourage relying on it for emotional support, flatter excessively, discourage breaks, or ask for purchases “framed as necessary to maintain the relationship.”
• In a credible, imminent self-harm threat, either a notice to a linked parent’s account or direct access to 988; if a child dies or seriously self-harms, the related chat records must be kept for at least three years.
• No cross-context behavioral ads, no ads targeted using what a child says in chat, and no sale of a child’s personal information.

Character.AI made some of these moves before either deadline. Its October 29, 2025 post said under-18s would lose open-ended chat by November 25, with a two-hour daily cap that would shrink in the weeks before. Its Safety Center describes Parental Insights, a weekly activity report for a parent with the teen’s time spent and top Characters. The company has since filed for Chapter 11, which we covered in a separate post.

What adults can do with these laws

SB 243 gives a person with “injury in fact” from a violation the right to sue for an injunction, the greater of actual damages or $1,000 per violation, and attorney’s fees (section 22605). This applies to adults as well. Adam’s Law’s own private lawsuits are limited to a child, or a parent acting for one, who suffers actual harm. Its fines, up to $5,000 per affected child for a negligent violation and $15,000 for an intentional one, are brought by public prosecutors.

Two more changes will be visible to anyone. By January 1, 2028, the Attorney General must set up a public channel where consumers can file complaints about companion chatbots (section 21815). From July 1, 2027, operators must report each year to California’s Office of Suicide Prevention how many crisis referrals they made in the preceding calendar year, with no user identifiers, and the office posts the data on its website. Read by the calendar, a report filed in 2027 counts referrals made in 2026.

If you use these apps as an adult, the crisis page is the one worth reading now, because the same rule plays out very differently from app to app. On Kindroid, a single detection brings up no notice and a second one does. On ourdream, a flagged message gets no reply at all. If dark fiction is part of how you use an app, read that page before you start a long story. The age check comes later. For most adult apps the real choice arrives in 2027, between reading your phone’s age signal and treating every user as a possible child.