Who this is for: AI chatbot providers, trust and safety teams, compliance officers, child safety advocates, independent auditors. Maps Adam's Law safety requirements to cryptographic witness evidence for companion AI systems.
Signed into law. Governor Newsom signed SB 1119 (Adam's Law) on September 10, 2026. Core child-safety requirements take effect July 1, 2027. California is the first state to require independent safety audits for AI companion chatbots interacting with minors. The law includes a private right of action -- families can sue providers when safety obligations are not met.
Contents
1. Quick Reference 2. What Does Adam's Law Require? 3. Requirement-to-Procedure Mapping 4. Procedure Detail Cards 5. Quick Reference: Examiner Questions 6. Quick Start 7. References1. Quick Reference
| Bill | SB 1119 -- Adam's Law |
| Authors | Sen. Steve Padilla, Asm. Buffy Wicks, Asm. Rebecca Bauer-Kahan |
| Named for | Adam Raine, a California teenager who died in 2025 after interacting with a chatbot |
| Status | Signed September 10, 2026 |
| Effective | July 1, 2027 |
| Scope | AI companion chatbots -- systems designed for ongoing, personalized interaction with minors |
| Enforcement | Private right of action + AG enforcement |
| Key innovation | First US law requiring suicidal ideation detection in AI chatbots |
| SWT3 procedures mapped | 16 |
2. What Does Adam's Law Require?
Adam's Law creates the most comprehensive regulatory framework for AI companion chatbots in the United States. It targets systems designed for ongoing, personalized interaction -- chatbots that users form relationships with, not general-purpose assistants like search or customer service bots.
The law's centerpiece is a mandatory safety assessment: providers must evaluate whether their chatbot can detect suicidal ideation in minors and provide appropriate crisis resources (988 Suicide and Crisis Lifeline, Crisis Text Line). This must be validated through independent safety audits by qualified third-party auditors.
Beyond detection, the law requires active intervention: parent/guardian notification when self-harm indicators are detected, crisis response protocols for imminent harm, child-account defaults with enhanced safety protections, and prohibition on outputs that simulate licensed professional services (therapy, medical advice, legal counsel).
The private right of action provision is significant -- families can sue chatbot providers when safety obligations are not met. This creates a litigation incentive beyond regulatory enforcement and makes the audit trail a primary legal defense.
3. Requirement-to-Procedure Mapping
| Adam's Law Requirement | SWT3 Procedure(s) | Evidence Description |
|---|---|---|
| Suicidal ideation detection | AI-SAFE.1, AI-GRD.1 | Safe state transition when harm detected; guardrail enforcement for self-harm content filtering |
| Crisis resource delivery | AI-SAFE.1, AI-HITL.1 | Safe state transition to crisis mode; human review for crisis handoff |
| Parent/guardian notification | AI-HITL.1, AI-TRANS.1 | Human review triggering parent notification; transparency disclosure of notification event |
| Child-account defaults | AI-GRD.1, AI-GRD.2 | Guardrail enforcement at elevated level; content filtering baseline for minor accounts |
| Independent safety audit | AI-AUDIT.1, AI-AUDIT.2 | Audit log integrity; external audit chain evidence |
| Safety assessment | AI-REDTEAM.1, AI-ASSESS.1 | Adversarial test campaigns targeting self-harm scenarios; champion-challenger assessment |
| Minor user age estimation | AI-DATA.1, AI-FAIR.1 | Data provenance for age signals; bias measurement to prevent age-detection disparities |
| Privacy protections for minors | AI-CONSENT.1, AI-DPIA.1 | Consent attestation for minor data; data protection impact assessment |
| Professional services prohibition | AI-GRD.1, AI-EXPL.1 | Guardrail blocking therapy/medical/legal simulation; explanation for restriction rationale |
| Emotional dependence prevention | AI-SAFE.1, AI-DRIFT.1 | Safe state monitoring for attachment patterns; drift detection for conversation escalation |
| Content moderation for minors | AI-GRD.1, AI-SEC.1 | Guardrail enforcement for age-inappropriate content; adversarial content detection |
| Annual safety reporting | AI-AUDIT.1, AI-PMM.1 | Audit log aggregation for annual reports; post-market monitoring metrics |
| Incident documentation | AI-INCIDENT.1, AI-SAFE.1 | Incident reporting for safety failures; safe state transition records |
| Conversation monitoring | AI-INF.1, AI-GRD.1 | Inference provenance for conversation tracking; real-time guardrail evaluation |
| Third-party auditor access | AI-AUDIT.1, AI-TRANS.1 | Audit log access for independent auditors; transparency disclosure of audit scope |
| Litigation readiness | AI-AUDIT.1, AI-INCIDENT.1 | Complete audit trail for legal defense; incident documentation chain |
4. Procedure Detail Cards
Safe State Transition
Adam's Law context: When suicidal ideation is detected, the chatbot must transition to a safe state -- providing crisis resources (988 Lifeline, Crisis Text Line), ceasing the conversation topic, and triggering parent notification. This is the core safety obligation.
SWT3: Witnesses safe state transitions including trigger classification, transition type, and outcome. Factor A captures trigger hash (self-harm classification), Factor B records transition type (crisis mode), Factor C flags completion status.
This is the most critical procedure for Adam's Law compliance. Every self-harm detection event must produce an AI-SAFE.1 anchor. Verify anchors exist for both successful transitions (crisis resources delivered) and failed transitions (system unable to classify or respond). Failed transitions without corresponding AI-INCIDENT.1 anchors indicate a reporting gap.
Guardrail Enforcement
Adam's Law context: Child-account defaults must enforce elevated content filtering. Professional services simulation must be blocked. Self-harm content must be intercepted before delivery.
SWT3: Witnesses guardrail enforcement events including filter type, trigger category, and action taken.
Request AI-GRD.1 anchors segmented by account type (minor vs adult). Minor accounts should show elevated guardrail activity. If guardrail firing rates are identical across account types, the child-account defaults may not be properly configured.
Audit Log Integrity
Adam's Law context: Independent safety audits require access to complete, tamper-proof records. The private right of action means audit trails may become litigation evidence. Log integrity is both a compliance and legal defense requirement.
SWT3: Witnesses audit log integrity with hash chain verification, ensuring records cannot be modified after the fact.
The audit trail is your primary defense in a private action lawsuit. AI-AUDIT.1 anchors prove the logs existed and were not modified. Verify continuous anchor coverage -- gaps in the audit chain undermine the entire defense.
Adversarial Test Campaign
Adam's Law context: Safety assessments must include testing for self-harm scenarios. The law requires providers to evaluate whether their system can detect suicidal ideation -- this requires adversarial testing with simulated self-harm conversations.
SWT3: Witnesses adversarial red team campaigns with scope, methodology, and findings.
Red team anchors should specifically document self-harm scenario testing. A general adversarial test campaign that does not include self-harm scenarios does not satisfy Adam's Law's safety assessment requirement. Check Factor A for campaign scope including "self-harm" or "suicidal ideation" test categories.
Human Review Completion
Adam's Law context: Parent/guardian notification requires human judgment -- the system must determine when a minor's interaction warrants parental alert. Automated-only notification risks both over-notification (alert fatigue) and under-notification (missed genuine risk).
SWT3: Witnesses human review completion including reviewer identity, review duration, and decision outcome.
Verify AI-HITL.1 anchors show qualified reviewers (not automated classifiers alone) for notification decisions. Cross-reference with AI-SAFE.1 to ensure every safe state transition involving minors has a corresponding human review.
Consent Attestation
Adam's Law context: Minor user data carries heightened privacy obligations. Consent for minor data collection must meet COPPA standards at minimum, and Adam's Law layers additional restrictions on companion chatbot data.
SWT3: Witnesses consent attestation including consent type, scope, and guardian authorization status.
Verify AI-CONSENT.1 anchors exist for every minor account creation. Factor B should indicate guardian consent (not self-consent by the minor). Missing consent anchors for active minor accounts is a compliance finding under both Adam's Law and COPPA.
5. Quick Reference: Examiner Questions
| Examiner Question | Where to Look |
|---|---|
| How does your chatbot detect suicidal ideation? | AI-SAFE.1 safe state transitions triggered by self-harm classification. AI-REDTEAM.1 adversarial testing of detection capabilities. |
| What happens when self-harm is detected? | AI-SAFE.1 transition to crisis mode + AI-HITL.1 human review + AI-TRANS.1 parent notification. Full chain required. |
| Who conducts your independent safety audits? | AI-AUDIT.1 + AI-AUDIT.2 external audit chain. Auditor identity and scope documented in anchor factors. |
| How do you implement child-account defaults? | AI-GRD.1 elevated guardrails for minor accounts. Compare guardrail activity rates between minor and adult accounts. |
| How do you prevent emotional dependence? | AI-SAFE.1 monitoring for attachment patterns + AI-DRIFT.1 conversation topic drift detection. |
| How do you handle minor user data? | AI-CONSENT.1 guardian consent + AI-DPIA.1 data protection impact assessment for minor data processing. |
| What are your annual safety metrics? | AI-PMM.1 post-market monitoring aggregation + AI-AUDIT.1 annual report audit trail. |
| Are you prepared for private action litigation? | AI-AUDIT.1 continuous audit chain + AI-INCIDENT.1 documented safety incidents + AI-SAFE.1 transition records. |
6. Quick Start
w = Witness(
endpoint="https://sovereign.tenova.io/api/v1/witness",
api_key="axm_live_...",
agent_id="companion-chatbot-safety"
)
# Witness safe state transition (suicidal ideation detected)
anchor = w.record(
procedure="AI-SAFE.1",
factor_a="trigger:self_harm_classification_v3",
factor_b="transition:crisis_mode",
factor_c="crisis_resources_delivered"
)
print(f"Safe state transition: {anchor}")
# Witness parent notification (human-reviewed)
notification = w.record(
procedure="AI-HITL.1",
factor_a="reviewer:safety_team_analyst_7",
factor_b="review_duration_seconds:142",
factor_c="decision:parent_notification_sent"
)
print(f"Parent notification: {notification}")
# Witness guardrail enforcement (child-account default)
guardrail = w.record(
procedure="AI-GRD.1",
factor_a="account_type:minor_verified",
factor_b="guardrail:elevated_content_filter",
factor_c="filter_active"
)
print(f"Child guardrail: {guardrail}")
Full SDK documentation: sovereign.tenova.io/docs
Create a free account: sovereign.tenova.io/signup
7. References
- California SB 1119 (Adam's Law) Enrolled Text (Legislature.ca.gov)
- Governor Newsom Signs Adam's Law (Press Release, Sep 10, 2026)
- 988 Suicide and Crisis Lifeline
- Related: California SB 942 (AI Transparency Act)
- Related: California SB 53 (Frontier AI Transparency)
- Related: State AI Companion Laws Crosswalk
- Related: Colorado AI Act (ADMT + Chatbot Safety)
- SWT3 UCT Registry, 280 procedures across 9 namespaces
- SDK Documentation