California Signs New Chatbot Child Safety Laws

By Kevin Fernando · Oct 7, 2026 · 11 min read

California has enacted Senate Bill 243 and Senate Bill 1119, requiring AI chatbot operators to implement age verification, content safety controls, data collection limits, and crisis intervention protocols for minors. SB 243 takes effect January 1, 2026, while SB 1119 phases in requirements beginning July 1, 2027. Violations carry civil penalties up to $15,000 per intentional violation.

California has become the first state to establish a comprehensive regulatory framework for AI chatbots used by minors. Governor Newsom signed Senate Bill 243 into law on October 13, 2025, creating binding obligations for chatbot operators to protect children from harmful content, manipulative design, and data exploitation. A follow-up law, Senate Bill 1119, enacted later that year, expanded these requirements further and introduced enforcement mechanisms including private rights of action.

These laws represent a shift in AI governance: rather than waiting for federal regulation or issuing broad principles, California has written specific operational requirements into statute. Developers must now verify that users are minors, filter certain categories of harmful content, limit how much personal data they collect, and detect and respond to signs of suicidal ideation. The requirements phase in over time, giving companies until January 1, 2026 to begin compliance and July 1, 2027 to implement the most complex obligations.

For researchers, policy analysts, and business stakeholders, understanding these laws means understanding what "child safety regulation" actually requires in practice, where the legal obligations are ambiguous, and how California's approach compares to emerging frameworks elsewhere.

What California's new chatbot child safety laws require

SB 243 and SB 1119 impose five core obligations on chatbot operators. First, operators must verify that a user is a minor before providing access to certain features or content. Second, they must filter or restrict content that could cause harm to minors, including content promoting self-harm, suicide, eating disorders, and illegal drug use. Third, they must limit the personal data they collect from minors and how long they retain it. Fourth, they must detect when a minor is expressing suicidal ideation and provide referrals to crisis services. Fifth, beginning July 1, 2027, operators must conduct risk assessments before launching new or substantially modified chatbots and document the safeguards they have put in place.

Some requirements take effect immediately upon signing, while others phase in by January 1, 2026 and July 1, 2027. The laws do not require operators to block minors from using chatbots entirely; instead, they require specific protective measures within the chatbot experience. A chatbot can continue to serve minors, but only if the operator has implemented the specified controls.

Which companies and systems are covered

SB 243 applies to any operator making companion chatbot platforms available to users in California. The law specifies no minimum company size or user threshold. A startup with a single chatbot and a small user base faces the same legal obligations as a large technology company with millions of users, provided the chatbot is accessible to California residents.

A "companion chatbot" under the law is an AI system designed to simulate ongoing relationships, provide emotional support, or engage in extended conversation with users. This covers systems marketed as friends, mentors, or emotional companions. The laws do not appear to cover transactional chatbots, those designed to answer questions, book appointments, or process orders, though the distinction can blur in practice. A chatbot that both answers questions and engages in relationship-building conversation would likely be covered.

The laws apply based on where the user is located, not where the company is based. If a chatbot is accessible to someone in California, the operator must comply with California law, even if the company is headquartered elsewhere or operates from outside the United States.

How developers must verify and protect minors

The laws require operators to implement multiple overlapping protections. These protections operate at different points in the user journey: at signup, during ongoing use, and in response to specific harmful content or behavior.

Age verification and access controls

Operators must verify that a user is a minor before providing access to the chatbot. SB 1119 specifies a method: operators must request age bracket data from the operating system or app store when the application is downloaded and launched. This means that if a user is downloading the chatbot through Apple's App Store or Google Play, the operator should request age information that the platform has already collected, rather than asking the user to self-report their age.

This approach reduces reliance on user self-reporting, which is easily circumvented by minors who misrepresent their age. However, it also means that compliance depends on the operating system provider or app store providing accurate age data. For web-based chatbots not distributed through app stores, the law is less explicit about which verification method satisfies the requirement, creating ambiguity that operators will need to navigate.

Once a user is identified as a minor, the operator must maintain that designation throughout the session and apply minor-specific protections to all interactions.

Content safety and filtering standards

Operators must filter or restrict chatbot responses that promote or encourage self-harm, suicide, eating disorders, or illegal drug use. The law does not require operators to block all discussion of these topics, a chatbot could provide factual information about eating disorders or direct a user to resources, but it prohibits responses that encourage or normalize these behaviors.

Beginning July 1, 2027, operators must also detect when a minor expresses suicidal ideation and issue a referral notification to a crisis service provider. The operator must then submit an annual report to the California Department of Public Health documenting how many such referrals were issued. This creates a dual obligation: the operator must both filter harmful chatbot outputs and monitor user inputs for signs of crisis.

The law does not specify the exact filtering mechanism or the precision required. Operators could use keyword matching, machine learning classifiers, or manual review, provided the method reasonably catches the covered harms. This flexibility allows different technical approaches but also leaves room for inconsistency across platforms.

Data collection and retention limits

Operators must limit the personal data they collect from minors and the duration for which they retain it. The law restricts collection to data necessary for the chatbot to function or to comply with legal obligations. Operators cannot collect data for secondary purposes like targeted advertising or behavioral profiling unless they have explicit consent from a parent or guardian.

Retention limits require operators to delete personal data from minors once it is no longer necessary for the stated purpose. If a chatbot collects conversation history to improve responses, the operator must delete that history after a defined period or when the user requests deletion. The law does not specify a maximum retention period, but it establishes that indefinite retention is not permissible.

These requirements create tension with common chatbot design practices. Many chatbots rely on long conversation histories to maintain context and personalization. Under California law, operators must either shorten retention windows, obtain parental consent for longer storage, or redesign their systems to function with less historical data.

Enforcement, penalties, and compliance timeline

SB 243 and SB 1119 are enforced through multiple mechanisms. The California Attorney General and local prosecutors can bring civil actions against operators for violations. Additionally, SB 1119 creates a private right of action, allowing individuals or parents to sue operators directly.

Penalties differ between the two laws. Under SB 1119, violations carry civil penalties of up to $5,000 per negligent violation and $15,000 per intentional violation. AB 1709, a related law addressing social media platforms, carries higher penalties: up to $50,000 per minor for knowing violations and up to $25,000 per minor for negligent violations. The difference reflects the legislature's judgment that social media platforms pose a higher risk to minors than chatbots, or that social media companies have greater resources to absorb penalties.

The compliance timeline is staggered. SB 243 takes effect on January 1, 2026, with certain requirements taking effect immediately upon signing and others phased in by July 1, 2027. This delay gives operators time to build new systems, though it also means that some requirements will not be tested until well after the law takes effect.

Companies that fail to comply face both civil liability and reputational risk. A single enforcement action or lawsuit could trigger media coverage and user backlash, particularly if the violation involved harm to a minor.

How this compares to federal and international approaches

California's framework is more prescriptive than the federal approach in the United States. The federal government has not enacted comprehensive AI regulation; instead, it relies on existing laws like the Children's Online Privacy Protection Act (COPPA) and the FTC Act, which address privacy and deceptive practices but do not mandate specific safety features like content filtering or crisis detection. President Trump's Executive Order 14179, issued January 23, 2025, directed the Administration to remove barriers to AI leadership rather than impose new restrictions.

In contrast, the European Union has taken a different approach. The EU AI Act places an outright ban on AI systems that include cognitive behavioral manipulation of people or vulnerable populations such as children. This is a prohibition on certain types of AI systems rather than a requirement for specific safety features. California's approach sits between these two poles: it does not ban chatbots but requires operators to implement specific protections.

Other U.S. states are moving toward regulation. In the 2025 legislative session, all 50 states introduced AI legislation, and thirty-eight states adopted or enacted around 100 measures. However, most of these measures address data privacy, algorithmic transparency, or sector-specific concerns rather than child safety in chatbots. California's SB 243 and SB 1119 are among the most detailed state-level frameworks for this specific use case.

The variation across jurisdictions creates compliance complexity for operators. A company serving users in California, the EU, and other U.S. states must navigate different rules: California's specific operational requirements, the EU's categorical bans, and varying state-level privacy laws. In practice, many operators will likely adopt California's standards as a baseline, since compliance with California often exceeds compliance with less stringent jurisdictions.

What this means for AI developers and platforms

These laws impose real development costs and product constraints. Operators must build or integrate age verification systems, content filtering models, data minimization architectures, and crisis detection pipelines. For small startups, these requirements may be prohibitively expensive; for large technology companies, they are manageable but still represent significant engineering investment.

The laws also constrain product design choices. A chatbot designed to build long-term relationships by remembering every detail of past conversations cannot do so with California minors without violating data retention limits. A chatbot trained to personalize responses based on behavioral patterns cannot do so without either obtaining parental consent or redesigning its learning pipeline. These constraints may reduce the perceived quality of the user experience, though they also reduce the risk of manipulation or harm.

For platforms that host third-party chatbots, the laws create liability questions. If a third-party developer builds a chatbot on a platform's infrastructure and that chatbot violates the law, is the platform operator liable? The law does not explicitly address this scenario, but the private right of action and broad definition of "operator" suggest that platforms could face liability for chatbots they host or enable. This may push platforms to implement stricter vetting and monitoring of third-party developers.

Product design changes developers may need to make

Developers may need to implement several concrete changes. First, they must add age verification at signup, likely by requesting age bracket data from the operating system rather than relying on user self-report. This adds a step to the signup flow and may reduce conversion rates if users find it intrusive.

Second, developers must implement content filtering that blocks or modifies chatbot responses containing certain harmful themes. This typically involves training or fine-tuning a machine learning model to classify responses before they are sent to the user. The filtering must be accurate enough to catch harmful content without blocking legitimate discussions, a difficult balance.

Third, developers must redesign data retention policies. Instead of storing all conversation history indefinitely, they must delete data after a defined period or when the user requests deletion. This may require changes to the database architecture and the conversation context system. A chatbot that relies on full conversation history for context may need to be redesigned to function with shorter or summarized histories.

Fourth, developers must implement crisis detection that identifies when a minor is expressing suicidal ideation and automatically generates a referral to a crisis service. This requires a separate machine learning model trained to detect crisis language and integration with crisis service providers. The referral mechanism must be reliable and must log each referral for annual reporting.

These changes are not technically insurmountable, but they require deliberate investment and careful design to avoid false positives (blocking legitimate content) or false negatives (missing harmful content).

Compliance challenges and open questions

The laws leave several ambiguities that operators will need to navigate. First, the definition of "companion chatbot" is not precise. A chatbot that sometimes provides emotional support and sometimes answers factual questions may or may not be covered. Operators will need to make judgment calls or seek legal guidance.

Second, the content filtering requirements are stated in general terms, "promote or encourage self-harm", but do not specify what language or themes trigger the filter. Operators must make decisions about edge cases: Does a chatbot discussing the causes of eating disorders "promote" eating disorders? Does a chatbot providing factual information about suicide methods "encourage" suicide? These judgments affect both compliance and user experience.

Third, the age verification requirement references the Digital Age Assurance Act, which requires requesting age bracket data from the operating system. However, not all chatbots are distributed through app stores, and not all operating systems provide age data. Web-based chatbots, for example, may not have access to this data. The law does not specify what alternative verification methods satisfy the requirement.

Fourth, the private right of action creates uncertainty about litigation risk. Who can sue? Can a parent sue on behalf of a minor? Can a minor sue directly? What damages are recoverable? These questions will likely be answered through litigation, creating a period of legal uncertainty for operators.

Finally, the laws require risk assessments beginning July 1, 2027.

What happens next in AI child safety regulation

California's laws are likely to influence regulation in other states and at the federal level. The state has historically set the pace for technology regulation, and other jurisdictions often follow. Thirty-eight states adopted or enacted around 100 AI measures in 2025, suggesting that appetite for AI regulation is broad. Some of these measures may be inspired by California's approach or may attempt to address similar concerns.

At the federal level, Congress has considered multiple AI regulation bills, though none have passed. California's specific framework may serve as a template for federal legislation or may prompt Congress to act preemptively to establish a national standard that preempts state-by-state regulation. Alternatively, the federal government may continue to rely on existing laws and agency oversight, leaving regulation to the states.

International developments will also shape the trajectory. The EU AI Act, now in effect, will influence how companies design systems globally. If EU requirements and California requirements diverge significantly, companies may face pressure to comply with the more stringent standard across all markets. Over time, regulatory convergence around a small number of frameworks is likely.

For chatbot operators, the immediate implication is that compliance with California law should be viewed as a baseline, not a ceiling. Additional requirements from other states, the federal government, or the EU are likely to follow. Building flexible, privacy-first architectures now will reduce the cost of adapting to future requirements.

Do this automatically

Verifying age brackets at signup, filtering responses across multiple harm categories during ongoing use, and auditing retention practices across thousands or millions of active users represents a continuous operational burden that grows with each new user and interaction. Software systems can automate these verification checkpoints, content filtering rules, and data lifecycle management tasks at scale. IMBeacon.ai runs this kind of analysis end to end and cites every figure back to its original source, across as many subjects as you need, with Mimir attaching a citation to every figure. It will not make the decision for you. It gathers and cites the evidence; the judgment stays yours, and deliberately so. Every plan includes it, including the free one.

Start your first report and run it on something you already know well.

This article is general information and research commentary. It is not professional advice, and it does not account for your particular circumstances.