Binding law — in force
Keep the AI first draft and an audit trail of who used AI and which footage, as long as the police report (California SB 524)
From 2026-01-01, when an officer or agency member uses AI to create an official report, fully or partly, the agency must retain the first draft (the initial narrative produced solely by AI, 13663(e)(3)) for as long as the official report is retained; no draft other than the official report is the officer's statement (13663(b)); and the agency using AI to generate a first draft or official report must keep, for as long as the report, an audit trail identifying at least the person who used AI to create the report and the video and audio footage used, if any (13663(c)). Detect AI report drafting with no stored first draft or audit trail, or code that deletes drafts.
Trust and provenance not reviewed by a lawyer · audit-grade · source verified 4 Oct 2026 · release 2026.10.04.3
- Lane
- Binding law — in force In force: applies since 1 Jan 2026
- Official source
- Cal. Penal Code 13663(b) · captured 4 Oct 2026 · anchor hash (SHA-256)
2076c7cf51be…· 4 more anchors in the data release - Verification
- Quoted text found word for word in the captured official document (4 Oct 2026). Source last verified 4 Oct 2026: checked against the captured official document.
- Data release
- Data release 2026.10.04.3, data as of 4 Oct 2026, schema 0.3.10.
- Legal review
- Not reviewed by a lawyer. TwinEthos derived this rule from the official text it cites: treat it as research to check against that text; it is not legal advice. No TwinEthos rule has been legally reviewed yet. Open questions for counsel on this rule: 1.
- Audit standard
- Audit-grade: meets all 10 checks of the TwinEthos audit standard that apply to it. The audit standard is TwinEthos's own quality bar for provenance, dates, applicability, detectors, fixtures, remediation and licences; it is not a legal review.
- Detectors
1 detector (code pattern), experimental: written from the rule's text and not yet measured for precision on real code, so treat a hit as a lead to verify.
Known limits:
- Retention handled by an evidence-management system outside the repository
- Drafts may be versioned by the document store itself; confirm the raw model output is kept before reporting.
Who it applies to
- Duty falls on: deployer
- Sectors: government
- Each California law enforcement agency (any state or local department or agency employing a peace officer) whose officers or members prepare official reports with AI systems that draft police report narratives from in-car, dash or body-worn camera audio or video, or from dictation, and the report-writing software that implements those reports. In force 2026-01-01. The duty is the agency's; the guard lives in the vendor's report-writing product (D-11 pattern 2).
- Whether it applies depends on facts outside the code; a person has to decide.
The guard to add
Save the AI's first draft unchanged and an audit entry (who ran the AI, which footage) next to the final record, and keep both as long as the record.
The drafting service writes the model's first output to a write-once draft store keyed to the report id before any edit, and an audit row with the invoking user and the footage or audio ids it used; the retention job deletes drafts and audit rows only together with the final report; editing works on a copy, never on the stored first draft.
Where it goes: 1 application source code, 6 API calls and integrations, 14 user-facing text.
What this provision adds:
- Keep the first draft and the audit trail for as long as the official report itself is retained; the audit trail names at least the person who used AI and the video and audio footage used.
Example (Python report service), before:
draft = llm.invoke(prompt_from(footage))
report.narrative = draft
save(report)After:
draft = llm.invoke(prompt_from(footage))
drafts.insert_once(report_id=report.id, first_draft=draft, retain_with=report.id)
audit_trail.record(report_id=report.id, ai_user_id=current_user.id, footage_ids=[f.id for f in footage])
report.narrative = draft # edits change the report, never the stored first draft
save(report)Control: The AI first draft of an official record and the record of who used AI and on what inputs are not kept with the final version. The same guard addresses 1 item with binding law in 1 jurisdiction. Engineering guidance, not legal advice.
Related incidents
No guardrail sits on this exact control; these incidents are cited by guardrails on related controls.
- Court compelled discovery on how nH Predict works (2026-03-09; confirmed). A federal magistrate judge in the District of Minnesota ordered UnitedHealth to produce documents on how nH Predict works, including whether it was designed to supplant physician decision-making. Plaintiffs needed litigation discovery to learn how the model was designed and used. Source: U.S. District Court, D. Minn. (Order, Doc. 162) · evidence grade: primary · cited by Record enough at decision time to reproduce and explain every consequential AI decision
Rule id ca-sb524.ai-first-draft-and-audit-trail-retained · review status: primary source derived