Executive Briefs

California's SB 1119 sets how companion chatbots behave with anyone under 18. On a school account, the district can't change the defaults.

Adam's Law locks protective defaults on child accounts, routes crisis notices to parents, adds audits, and lets prosecutors seek $15,000 per affected child.

Filed under
AI in Schools · California · Student Privacy
Published
Oct 29, 2026, 11:16:41 AM
Read
Statute

Cal. Bus. & Prof. Code Division 8, new Chapter 11.6, §§ 21810 to 21818 ("Adam's Law"); § 22602 amended · Ch. 190, Stats. 2026

Effective

Statute effective January 1, 2027, including age assurance (§ 21811), the Attorney General reporting channel (§ 21815), and enforcement (§ 21816). Child safety duties, record preservation, and advertising and data limits (§§ 21812, 21812.5, 21813) operative July 1, 2027. Interface testing with children and parents due January 1, 2028. First independent child safety audit due January 1, 2029 (or before first release, if later); January 1, 2032 for operators under $500 million in prior-calendar-year gross revenue, which are exempt from the audit section until that date.

Applies to

Any person that makes a companion chatbot, as defined in § 22601, available to users in California. The full duty set applies to operators that allow child users.

Last verified

October 27, 2026

The short version

  • SB 1119 is a product-design law, not a disclosure law. It tells companion chatbot operators how the product has to behave with anyone under 18: no memory across sessions by default, no push notifications, a one-hour session cap, and a two-hour daily cap, all changeable only by a linked parent.
  • The definition turns on what the chatbot can do, not how it's marketed. A system "capable of meeting a user's social needs" is in scope. Education isn't one of the listed exclusions, and only colleges, not K-12, get a carve-out for educational use.
  • An operator that won't determine age has to treat every user as a child. § 21811 gives two paths: determine age through California's age-signal framework, or apply the child protections to all users.
  • Enforcement runs on two tracks. Public prosecutors can seek up to $5,000 per affected child for a negligent violation and $15,000 for an intentional one. A harmed child or parent can sue for actual damages and attorney's fees.
  • Independent audits are new. They must be done by a state-registered AI auditor under AB 1405, starting January 1, 2029, or January 1, 2032 for operators under $500 million in prior-calendar-year gross revenue, which are exempt from the audit section until that date.
  • There's a six-month gap to watch. The bill deletes the existing minor-specific chatbot duties in § 22602, apparently from January 1, 2027, while the replacement duties start July 1, 2027.

$15,000 per affected child. That's the ceiling on a civil penalty for each intentional violation of Adam's Law, and the count runs per affected child for each violation. A negligent violation carries up to $5,000 per affected child. Both are available to the Attorney General and other public prosecutors under § 21816(a).

On September 10, 2026, Governor Newsom signed SB 1119 (Padilla, Wicks, Bauer-Kahan) as Chapter 190, Statutes of 2026. The bill passed the Assembly 69-4, and the Senate concurred 40-0. It's named for Adam Raine, a California teenager who died in 2025, and it adds a new Chapter 11.6 to the Business and Professions Code titled "Adam's Law."

SB 1119 does three things. It replaces the minor-specific rules California adopted last year in SB 243 with a much larger set of design duties for any companion chatbot that allows child users. It requires every operator, whether or not it allows children, to handle age through a defined path. And it adds independent audits, a public incident-reporting channel run by the Attorney General, and civil enforcement by both prosecutors and families.

What changed

California's first companion chatbot law, SB 243 (Chapter 677, Statutes of 2025), took effect January 1, 2026. It required every operator to disclose that a chatbot isn't human where a reasonable person could be misled, and to maintain a published protocol against producing suicidal ideation, suicide, or self-harm content. For users the operator knew were minors, it added three duties: disclose that the user is talking to AI, remind the user to take a break every three hours, and take reasonable measures against sexually explicit content.

SB 1119 deletes those three minor-specific duties from § 22602 and replaces them with the new chapter. The general duties in § 22602(a) and (b) stay in force for all users, and SB 243's private right of action in § 22605 stays on the books.

The new chapter is organized around five obligations.

Age. § 21811 requires each operator to either determine a user's age through the Digital Age Assurance Act framework (Civil Code Title 1.81.9), with a fallback to the standard in Health and Safety Code § 27001, or apply the child protections to every user. An operator that excludes children has to publish a high-level description of how it does that (§ 21812(b)).

Risk assessment before release. Before making a new or substantially modified companion chatbot available, the operator has to perform and document a risk assessment covering each "covered harm" (physical or financial harm, severe psychological or emotional harm, a highly offensive privacy intrusion, and unlawful discrimination), cite the benchmarks and research it used, and document the measures it took to mitigate what it found (§ 21812(a)).

Design duties for child users. § 21812(d) requires a documented crisis response protocol with a clear referral to crisis services, and, when the operator determines there's a credible and imminent risk of self-harm, either notice to a linked parent or streamlined access to 988 on the child's device. It sets protective defaults that only a linked parent can change: persistent conversational memory off, push notifications off, a one-hour cap on a continuous session, and a two-hour cap per day. Parental controls must include the ability to disable access entirely for a child under 16. AI-identity notices have to repeat during extended interactions. And the operator must take reasonable measures to prevent fourteen listed behaviors, including encouraging self-harm, simulating romantic interest, claiming to be sentient or human, encouraging reliance on the chatbot for emotional support, using flattery disproportionate to the context, discouraging breaks, and encouraging a child to circumvent parental controls.

Advertising and data. § 21813 bars cross-context behavioral advertising to a child and advertising targeted using what a child says in a conversation, permits labeled contextual advertising, bars the sale of a child user's personal information gathered through the chatbot, limits use to what's necessary to provide the service, maintain safety, or comply with law, and bars dark patterns in feature and control design.

Accountability. § 21812.5 requires the operator to preserve conversation records indicating serious self-harm risk for at least three years after a credible-risk notice or after learning a child died or seriously harmed themselves. § 21814 requires independent child safety audits on a two-year cycle. § 21815 requires the Attorney General to stand up a public incident-reporting mechanism by January 1, 2028.

Where it actually reaches

The reach test is easiest to trace by asking three questions in order.

Is the product a companion chatbot? SB 1119 doesn't write its own definition. § 21810.5(f) adopts the SB 243 definition in § 22601: an AI system with a natural language interface that gives 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. Three products are excluded: bots used only for customer service, a business's operational purposes, productivity, and analysis related to source information, internal research, or technical assistance; video game bots limited to game topics; and stand-alone voice assistants that don't sustain a relationship across interactions. The test is capability. A general-purpose assistant that can carry an ongoing conversation is within reach of it. So is a tutoring or learning-companion product with a persona and memory across sessions. A tool that only answers questions about source material sits closer to the first exclusion.

Is the operator covered, and for what? An operator is "a person who makes a companion chatbot available to a user in the state." Two operators are carved out: a postsecondary institution offering a chatbot exclusively for educational use, and an employer offering one exclusively to its own personnel. There's no K-12 equivalent. Every operator has to meet the age requirement in § 21811 and, if it excludes children, publish how. The full duty set in §§ 21812 to 21816 applies to operators that allow child users once age is determined. An operator that doesn't determine age takes the second path in § 21811 and applies the child protections to all users, with the defaults locked unless it has actual knowledge a user is an adult.

Which duties land on student use? For a district, the practical question is what happens when a student uses a covered chatbot on a school account. Unless a parent account is linked, the defaults can't be changed (§ 21812(d)(3)(D)). The session cap and the daily cap hold, memory across sessions may be off by default (especially for students under 16), and the district has no authority under the statute to lift them. A crisis notification under § 21812(d)(1) runs to a linked parent. Nothing in the chapter routes it to the school.

That last point is where the district inventory matters. In one EdTech audit we ran, the district had roughly 60 tools on its own records and just under 500 reachable through its single sign-on. A district can't tell which of its students' tools are companion chatbots from the approved list alone, because most of what's reachable never went through approval.

What it costs to be wrong

Public enforcement. Under § 21816(a), the Attorney General and the other public prosecutors named in § 17204 can sue for a civil penalty of up to $5,000 per affected child for each negligent violation and up to $15,000 per affected child for each intentional violation, plus attorney's fees. Injunctive and declaratory relief are available to the Attorney General. The per-child count is what makes the exposure scale with the user base.

Private enforcement. Under § 21816(b), a child who suffers actual harm from a violation of the crisis-protocol, safeguard, default-setting, AI-notice, or preventive-measure duties in § 21812(d)(1) to (5), or a parent acting for that child, can sue the operator for actual damages, attorney's fees and costs, and injunctive or declaratory relief. Financial harm has to exceed $1,000 per child, and emotional harm has to amount to serious emotional distress. A violation of the chapter can't be the basis for a UCL action (§ 21816(c)). SB 243's separate action under § 22605, the greater of actual damages or $1,000 per violation, continues for the duties that remain in § 22602.

Independent audits. Starting January 1, 2029, audits must be done by a state-registered AI auditor under AB 1405, which sets the independence standards (§ 21814). The operator files a summary with the Attorney General within 30 business days of receiving the audit and posts a high-level summary on its website within 90 days. The Attorney General can obtain the full report for cause. The full report is confidential, but it's available for enforcement.

For vendors, the exposure is enforcement plus procurement. A California district that asks at renewal whether a product is a companion chatbot, which § 21811 path the vendor took, and where its child safety policy is published will get a document or an absence. From 2029, the audit summary becomes the artifact districts ask for. A vendor that can't produce a child safety policy is telling the district it either doesn't allow children or hasn't complied.

For districts, the exposure is operational and reputational rather than statutory, on the reading in this brief. A district that deploys a chatbot to students without knowing whether it's covered can't tell parents what protections apply. A district whose instructional plan assumes a chatbot remembers students across sessions may find that the default limits it. And a district whose suicide prevention policy under Education Code § 215 assumes the school hears about risk first should know that the statute routes the operator's notice to the parent.

Questions for your cabinet

Districts should raise these at the next cabinet meeting and again at the next board meeting:

  1. Which AI tools can our students reach today through our single sign-on, our managed devices, and our approved apps? Which of them hold a conversation, keep a persona, or remember a student across sessions? Those are the candidates for companion-chatbot coverage.
  2. For each candidate, which § 21811 path has the vendor taken: determining age, or applying child protections to every user? Where's its published child safety policy?
  3. Does any instructional use we've approved assume the chatbot remembers students across sessions, or runs longer than an hour at a time? Under Adam's Law, on a student account without a linked parent, it may not run longer than an hour, and memory across sessions may be off by default, especially for students under 16.
  4. The six-month gap. Between January 1 and July 1, 2027, the existing minor-specific chatbot duties come off the books and the new ones aren't yet operative. What does our own acceptable-use policy require of chatbots our students use during that window?
  5. When a chatbot operator notifies a parent of a credible self-harm risk, how does that reach our counseling staff, if at all? Does our Education Code § 215 suicide prevention policy account for risk identified outside the school?
  6. Does our DPA template, or our AB 1159 attestation, ask whether a product is a companion chatbot under § 22601? If not, who adds the question, and before which renewal?

Compliance path

What the law requires

From January 1, 2027, every operator of a companion chatbot available in California has to either determine users' ages or treat every user as a child. The child protections the second path applies don't become operative until July 1, 2027. From July 1, 2027, operators that allow children must assess risk before each release, publish a child safety policy, run a documented crisis protocol, ship the protective defaults locked to parent control, prevent the fourteen listed behaviors, keep advertising contextual and labeled, and preserve serious-incident records for three years. They must test their controls with children and parents by January 1, 2028, and complete an independent child safety audit by January 1, 2029 (January 1, 2032 for operators under $500 million in prior-calendar-year gross revenue, which are exempt from the audit section until that date).

What that means for districts

Build a list of the AI tools students can actually reach, not only the ones you approved, and sort it by whether each holds a conversation or a relationship. For every tool on that list, ask the vendor in writing whether the product is a companion chatbot under § 22601, which § 21811 path it uses, and where its child safety policy lives. Review any instructional design that depends on memory or long sessions before July 1, 2027. Decide what your acceptable-use policy requires during the January-to-July 2027 window. And connect your suicide prevention policy to the fact that operator notices go to parents.

What that means for vendors

If your product can carry an ongoing conversation with a student, assume a district will ask whether it's a companion chatbot, and have a written answer that maps to the § 22601 exclusions rather than to your marketing. Decide your § 21811 path early, because the second path locks the defaults for every user you can't confirm is an adult. A published child safety policy, a documented pre-release risk assessment, and an audit-ready control set are what the next California procurement cycle will check. Products sold to K-12 don't share the postsecondary carve-out.

What we do about it

We run EdTech audits that start from what students can actually reach through single sign-on, not from the approved list. For California districts, the audit now flags every tool that holds a conversation, keeps a persona, or remembers a user, and records the vendor's written answer on companion-chatbot status, its § 21811 path, and its child safety policy. Districts in DPA review pick up the same three questions alongside the AB 1159 attestation module, without a scope change.

Beyond your district

SB 1119 is the flagship of the thirteen bills Governor Newsom signed on September 10, 2026, and the package splits AI regulation along two lines. AB 1159 governs what student data can go into a model. SB 1119 governs how a companion chatbot behaves with a child. SB 867 bans companion chatbots in toys through 2030. AB 1856 addresses the age-signal framework that SB 1119's age path relies on. A product sold to California schools can sit under several of these at once, and compliance with one doesn't establish compliance with another.

The audit requirement is the piece most likely to travel. Other states have passed companion chatbot laws built on disclosure and crisis protocols, but SB 1119 pairs design duties with independent third-party audits and a state-run incident channel. Expect the audit summary to become a standard procurement artifact well beyond California, the same way California's original student privacy act became the template for other states' operator laws.

Crosswalk

Against SB 243. SB 243's general duties survive for every user: the AI-identity notice where a reasonable person could be misled (§ 22602(a)), the published self-harm protocol with crisis referrals (§ 22602(b)), annual reporting to the Office of Suicide Prevention beginning July 1, 2027 (§ 22603), and the § 22605 private right of action. Its three minor-specific duties are deleted. Adam's Law replaces the three-hour break reminder with a one-hour session cap and a two-hour daily cap, and replaces the sexual-content measure with a broader list of prohibited behaviors.

Against AB 1159. AB 1159 (Ch. 182, Stats. 2026) bars K-12, preschool, and higher-education operators from training or developing AI on covered student information. SB 1119 doesn't address training data. An AI tutor sold to a California district can be under both: AB 1159 for what it does with student information, SB 1119 for how it behaves in the conversation.

Against federal statute. COPPA covers operators of services directed at children under 13, or with actual knowledge of collecting from them, and centers on notice and parental consent for data collection. FERPA governs education records held by schools. Neither sets product-behavior duties for conversational AI. SB 1119 reaches every user under 18 and regulates the conversation itself.

Forward signal

Two open questions we're watching.

The first is how the age signal behaves on a school-managed device. § 21811's first path relies on the age-bracket signal the Digital Age Assurance Act requires operating system providers and app stores to send. On a district-managed device, the account holder, the age entered at setup, and who can link a parent account are all district configuration decisions. Whether a student on a managed device is signaled as a child, and whether a parent can ever be linked to a school-provisioned account, will decide whether the defaults can be changed for instructional use at all.

The second is the K-12 gap in the operator exclusions. The Legislature carved out postsecondary institutions offering chatbots exclusively for educational use and didn't carve out K-12 districts. Whether that silence means a district that builds or deploys its own chatbot is an operator, or simply that public entities were never "persons" under the chapter, isn't settled on the face of the bill. Districts building their own chatbots should get advice on it before July 1, 2027.

What to do this week

California districts: book a scoping call to identify which AI tools your students can reach through single sign-on, which of them are candidates for companion-chatbot coverage under SB 1119, and what each vendor's written position is on age, defaults, and child safety policy. We return a prioritized list sorted against the July 1, 2027 operative date.

Vendors serving California K-12: book a scoping call to work through whether your product is a companion chatbot under § 22601, which § 21811 path fits your user base, and the child safety documentation your district customers will ask for at the next renewal.

Sources and a note on what this is

Read against the chaptered text of SB 1119 (Chapter 190, Statutes of 2026, signed September 10, 2026), last verified October 27, 2026.

Statute tags for update tracking: CA.BPC.21810 (Adam's Law, Ch. 11.6) · CA.BPC.21811 · CA.BPC.21812 · CA.BPC.21814 · CA.BPC.21816 · CA.BPC.22601 (definition) · CA.BPC.22602 · CA.BPC.22605 · CA.SB1119.2025-26 · CA.CH190.2026 · CA.AB1405.2025-26 · CA.SB867.2025-26

This brief is general information about California law as of the verification date above. It isn't legal advice, and reading it doesn't create an attorney-client or advisory relationship. Statutes are amended and interpreted after enactment, and how SB 1119 applies to a particular district, product, or user depends on facts this brief doesn't address. Consult your own counsel before acting on anything here.

Claire Smizer-Muldoon
See clearly.
Act strategically.
Protect what matters.
Book a call
See if we're a fit.
Twenty minutes with Trish and Claire. Bring what's actually worrying you.
Start the Conversation →