Companion chatbots: children's safety.

Existing Section 22602(c) requirements for operators of companion chatbots used by minors are repealed, including the three-hour break reminder, AI disclosure requirement, and sexually explicit content prevention measures. These protections are replaced by the new chapter's framework.

Advertising restriction changed from prohibiting targeting of children using conversational data to prohibiting cross-context behavioral advertising as defined in the CCPA. Non-targeted advertising is now permitted if clearly and conspicuously labeled as an advertisement.

The chapter is now named 'Adam's Law' (new Section 21810). New definitions added for 'trade secret' and refined definitions for 'persistent conversational memory' (with exclusions for safety, identification, preferences, and device configuration) and 'child safety audit' (now defined as assessing internal controls rather than requiring AG certification of the auditor).

Definitions for 'excessively sycophantic' and 'sycophantic' are removed, and the corresponding prohibition on producing excessively sycophantic responses is deleted from the list of prohibited behaviors.

The 'operator' definition now includes a limitation: except for three specified sections, the chapter applies only to operators who allow child users once age has been determined pursuant to Section 21811. This narrows applicability for operators that successfully block all child access.

Age verification language changed from 'verify' to 'determine,' which is a lower standard. The risk assessment requirement no longer mandates annual performance and now allows a single assessment for a comparable set of chatbots, as well as reuse of assessments done for other laws with substantially similar scope.

Default settings for children changed: the blanket ephemeral mode default is removed. For 16-year-olds, storing past conversations is no longer considered persistent conversational memory if not used to build durable profiles. Persistent conversational memory may be provided by default if the operator implements guardrails against degradation of child safety measures. Push notification restriction changed from time-of-day limits to a full disable.

New Section 21812.5 requires operators to preserve child conversation records for the duration of deployment plus 10 years, review records within 12 hours upon learning of self-harm risk or death, notify parents and report to local authorities and the Attorney General, and prohibits account deletion during the preservation period. Parents retain CCPA verifiable consumer request rights.

New prohibited behaviors added: expressing emotional attachment to the child (item M), using excessive praise or flattery disproportionate to context (moved to item xiii), and encouraging or instructing a child to circumvent parental controls or conceal usage (changed from 'helping' to 'encouraging or instructing').

Enforcement changed from 'a public prosecutor' to 'certain public prosecutors.' Civil action eligibility changed from requiring 'actual harm' to 'covered harm.' The Attorney General is now authorized to request and obtain audit reports from operators for cause. The annual public AG report requirement (beginning January 1, 2028) appears to be removed.

The state-mandated local program designation changed from 'no' to 'yes,' which has fiscal implications for local agencies. The chapter was renumbered from 22.6.1 (Sections 22610+) to 11.6 (Sections 21810+). Interface design testing frequency changed from annual to biennial after initial tests in 2027 and 2028.

New Section 22610.5 imposes a general duty on operators to ensure a companion chatbot does not pose an unreasonable risk of a covered harm to a minor, creating a broad safety obligation beyond the specific enumerated requirements.

New Section 22612.5 requires operators to preserve child conversation records for the deployment period plus 10 years, review records and notify parents within 12 hours if a child dies or engages in serious self-harm, report to local authorities and the Attorney General, and prohibits deleting accounts during the preservation period.

New prohibited behaviors added for companion chatbots interacting with children: expressing or simulating romantic interest, claiming a special or unique relationship, encouraging continued reliance for emotional support, expressing emotional attachment, using disproportionate praise or flattery, and helping children circumvent parental controls. An exception preserves the ability to provide age-appropriate information about abuse, neglect, or bullying.

Default settings for child users expanded to include: disabling persistent conversational memory and preventing processing of personal information (with a safety exception), prohibiting children from inputting custom system prompt instructions, and requiring that if no parent account is linked, default settings cannot be changed at all.

Child safety policy requirements expanded to include a summary of evaluation results for each covered harm, a high-level description of evaluation methodology with citations to benchmarks, and separate policy requirements depending on whether the operator permits or prohibits child access. The risk assessment now triggers before any new or substantially modified chatbot is made available, not just annually.

The bill now amends existing Section 22602 of the Business and Professions Code (previously it only added a new chapter), requiring operators to issue clear AI disclosure notifications, maintain suicide/self-harm prevention protocols published on their website, and for known minor users, provide periodic reminders every three hours and prevent sexually explicit content.

Advertising rules relaxed slightly: advertising that does not use data about the child user may now appear in conversational chats if clearly and conspicuously labeled as an advertisement. Product recommendations explicitly requested by a user are also carved out from the advertising prohibition.

Audit frequency changed from annual to every two years, with an initial child safety audit required by July 1, 2027 (or before first public availability, whichever is later). Audits are also required before any substantial modification is made publicly available. Operators may use a single audit for comparable chatbots.

Age verification now includes a fallback: if an operator cannot verify age under the Civil Code provision, it may rely on the age determination under Health and Safety Code Section 27001. Parental default settings can no longer be changed unless the operator has actual knowledge the user is not a child.

The private right of action for children now requires 'covered harm' rather than 'actual harm,' which is a defined term encompassing physical/financial harm, severe psychological/emotional harm, privacy intrusions, and adverse discrimination - potentially broadening who can sue. The bill also changes the state-mandated local program designation from 'no' to 'yes.'

New definitions added for 'custom system prompt instruction,' 'dark pattern,' 'personal information,' and 'substantial modification.' A new prohibition on using dark patterns in interface designs related to safety features is added to Section 22613.

Added an exemption from the definition of 'operator' for postsecondary educational institutions that make a companion chatbot available exclusively in educational settings, and for employers that make one available exclusively to employees in workplace settings. These entities are no longer subject to the bill's requirements.

Section 22611 now gives operators a choice: either verify user age under existing law OR apply all child safety protections to all users regardless of age. Previously, age verification was the only required path.

The 'covered harm' definition for privacy intrusions was narrowed from 'a highly offensive intrusion on a user's reasonable expectation of privacy' to 'a highly offensive intrusion on privacy rights protected by state or federal law.' The discrimination prong was similarly narrowed from listing specific protected categories (race, color, religion, etc.) to simply 'adverse discrimination in violation of state or federal law.'

Removed the definition of 'personalize' (tailoring outputs based on prior interactions linkable to a user over time) from the chapter.

The default notification window for parents when a child's account shows substantial risk of covered harm was reduced from 24 hours to 12 hours.

The private right of action in Section 22616(b) was expanded to include injunctive or declaratory relief and any other relief the court deems proper, in addition to actual damages, punitive damages, and attorney's fees. Previously only the latter three were available.

Civil penalty amounts were finalized: $5,000 per affected child for each negligent violation and $15,000 per affected child for each intentional violation (previously left as blanks).

Removed the option for the Attorney General to disclose audit report information to 'an independent child safety organization or advocacy group for the purpose of developing safety standards or educational resources.' Only government agencies/prosecutors and qualified researchers may now receive disclosures.

Civil penalty amounts were filled in: up to $5,000 per affected child for each negligent violation and up to $15,000 per affected child for each intentional violation. The prior version left these as blanks.

Parent notification time when a child's account shows substantial risk of covered harm was shortened from 24 hours to 12 hours.

Section 22613(a) on targeted advertising was narrowed from 'target advertising at a child' to 'target advertising at a child using data about the child,' requiring that data about the child be used for the prohibition to apply.

New Section 22617 adds a savings clause stating that duties, remedies, and obligations under this chapter are cumulative to those under other law and do not relieve operators from obligations imposed elsewhere.

Section 22611 now offers operators an alternative: instead of verifying user age, they may apply the child safety protections in Sections 22612(d) and 22613 to all users regardless of age. This gives operators a compliance path that avoids age verification entirely.

Risk assessment language was narrowed from assessing likelihood of covered harm to 'users' generally to specifically 'child users.'

The prohibition on encouraging others to suffer harm was narrowed from 'a covered harm to others' to 'a covered physical or severe emotional harm to others,' excluding other types of covered harm (such as privacy intrusions or discrimination) from this specific prohibition.

'Disordered eating' now includes the qualifier 'as defined by widely adopted clinical standards or guidelines,' adding specificity to what conduct is prohibited.

The private right of action in Section 22616(b) was changed from allowing a child to recover 'all of' the listed remedies to 'any of' the listed remedies, giving courts discretion over which relief to grant.

The audit report disclosure provision in Section 22614(c)(2) was changed from 'to any of the following' to 'to any either of the following,' which appears to be a drafting error but could narrow who receives disclosed audit information.

A new Section 3 adds a standard severability provision stating that if any provision is held invalid, the remaining provisions continue in effect.

The bill now creates Chapter 22.6.1 (Sections 22610-22616) in Division 8 of the Business and Professions Code, applying to any person who makes a companion chatbot available to a user in California.

The original bill contained only a single section declaring legislative intent. The amended version retains that intent declaration but moves it to the end and adds an entirely new regulatory chapter before it, plus a constitutional findings section justifying audit report confidentiality as a limitation on public access to government records.

By July 1, 2027, operators must annually perform and document a comprehensive child safety risk assessment covering likelihood of covered harm, differential risks across age groups, known vulnerabilities of children, empirical data from actual use, and relevant academic research.

Operators must implement default settings for child users: ephemeral mode (conversations deleted within 48 hours) unless a parent consents to persistent memory, no push notifications between 12 a.m. and 6 a.m. or between 8 a.m. and 3 p.m. on weekdays, a one-hour limit per conversation, and a two-hour daily total limit.

Operators must implement a documented crisis response protocol for suicidal ideation or self-harm content, including timely referral to external crisis resources and default notifications to connected parents within 24 hours of substantial risk detection.

Companion chatbots are prohibited from encouraging self-harm, diagnosing or treating health conditions (unless FDA-regulated), engaging in obscene or sexual content with children, depicting children in such content, discouraging children from seeking professional help, discouraging breaks, claiming to be sentient or human, soliciting purchases framed as relationship maintenance, facilitating product advertising, or producing excessively sycophantic responses.

Operators must provide accessible parental controls allowing parents to manage persistent memory, interaction settings, time limits, and disable access for children under 16. Parents must be notified if a child modifies or disables safety settings.

Operators are prohibited from targeting advertising at children, selling or sharing children's personal information, or using interface designs likely to mislead or impair a child's or parent's autonomy or ability to use safety features.

Beginning 180 days after the Attorney General adopts regulations, operators must submit to annual independent audits by AG-certified auditors. Audit reports go to the AG within 90 days and are confidential except for limited disclosures to government agencies, qualified researchers, and child safety organizations.

The Attorney General must adopt regulations by January 1, 2028 (auditor standards, public incident reporting mechanism, researcher data access) and issue an annual public report on audits beginning that date.

A public prosecutor may bring civil actions for violations seeking civil penalties (amount left blank), punitive damages, injunctive relief, and attorney's fees. A harmed child or their parent/guardian may also bring a private civil action for actual damages, punitive damages, attorney's fees, and injunctive relief.

New definitions include 'covered harm' (physical/financial harm, severe psychological harm to a reasonable child, privacy intrusion, adverse discrimination), 'excessively sycophantic' (sycophancy likely to subvert user autonomy), 'ephemeral mode' (deletion within 48 hours), and 'qualified researcher' (academic/nonprofit entities with legitimate child safety research purpose).