Living dossier · 13 linked objects · graph state read October 1, 2026, 8:00 PM PT · last editorial review October 1, 2026 · original snapshot September 18, 2026 (archived below)

Healthcare AI governance

California’s 2025–26 session closed on September 30, 2026 with the health-AI package split in two. AB 1979 (a licensed clinician’s professional judgment governs when clinical decision tools inform care) and SB 503 (developer and deployer duties to look for foreseeable bias) are law, operative January 1, 2027†. SB 813 and AB 1405 (voluntary independent verification; an auditor registry) were signed September 9. AB 2575, which would have protected clinicians who override an algorithm, and SB 903, on AI and the mental-health workforce, were vetoed. The result recognizes clinical judgment without protecting the clinician who exercises it against the machine. WHO/Europe’s evidence that governance readiness, not deployment speed, is the binding constraint stands as a convened expert report, not WHO policy.

Session closed September 30, 2026 · final outcomes recorded October 1, 2026

What became law, by layer

Four layers that public coverage compresses into “California AI law”, each with its own final action. Current analysis: California’s Health-AI Experiment Is Now Law.

Swipe the table sideways for the remaining columns

Every AI measure in the final ledger, by the layer it governs. Signed first, then vetoed. Generated from the registry on October 1, 2026.
Bill	Layer	Title	Final action	Operative
AB 1331	Workplace surveillance	Workplace surveillance: bathrooms	SIGNED SEPTEMBER 30, 2026	Jan. 1, 2027†
AB 1883	Workplace surveillance tools	Workplace surveillance tools	SIGNED SEPTEMBER 30, 2026	Jan. 1, 2027†
AB 1979	Clinical decision support · professional judgment	Health care services: artificial intelligence	SIGNED SEPTEMBER 30, 2026	Jan. 1, 2027†
SB 503	Developer and deployer bias duties	Health care services: artificial intelligence	SIGNED SEPTEMBER 30, 2026	Jan. 1, 2027†
SB 947	Automated employment decision systems	Employment: automated decision systems	SIGNED SEPTEMBER 30, 2026	Jan. 1, 2027†
SB 951	AI-caused layoffs: notice	AI-caused layoffs: notice	SIGNED SEPTEMBER 30, 2026	Jan. 1, 2027†
AB 1405	AI auditor registry	AI auditors registry	SIGNED SEPTEMBER 9, 2026	Per text
SB 813	Independent verification organizations (voluntary)	AI: independent verification organizations	SIGNED SEPTEMBER 9, 2026	Per text
AB 2575	Clinician override protection	Health care services: artificial intelligence	VETOED SEPTEMBER 30, 2026	—
SB 903	Mental-health workforce	Mental health professionals: artificial intelligence	VETOED SEPTEMBER 30, 2026	—

† Presumptive operative date under California’s default rule; confirmed against each chaptered text as it is read. The veto messages for AB 2575 and SB 903 are posted as image-only documents; their stated reasons are reported once transcribed.

Archived snapshot — originally current through September 18, 2026, 3:00 PM PTPreserved as originally published. Superseded by the final status above; lifecycle language, deadlines and “pending” states in this section describe the record as it stood on that date, not the current state. This update note is the pre-deadline record.

What the law provides

Medical-record content and retention are governed by state record requirements and electronic-record certification criteria; neither requires that machine-generated clinical text be identifiable as such. FDA’s clinical-decision-support policy addresses device status, not authorship. Attestation law assumes a human author.

The gap

No instrument requires provenance capture at generation, model-version recoverability, or a face-of-the-record distinction between physician and machine text. Every surveyed instrument ends at the clinician, because the clinician is the only actor already licensed, insured and disciplinable.

The KPSGILL position

Require machine-readable provenance for AI-generated clinical text at the point of generation, and move physician responsibility from authorship to clinically material adoption. The twelve-dimension readiness framework treats validation, provenance, bias monitoring, accountability allocation and model change control as separate questions.

Full treatment →

The open question

Responsibility should follow control, and a clinician using a system she cannot inspect does not control it. Every instrument surveyed nonetheless lands responsibility on her. What allocation would survive a case where the model was right, the clinician overrode it, and the patient was harmed?

Litigation

No docket yet. The KPSGILL forecast is that contested board discipline over AI-assisted decisions arrives before any board publishes risk-tiered guidance.

Tracked legislation

Final actions are in the table at the top of this dossier; the register table below is the archived September state.

From the California bill register. None of these is law.
Bill	Subject	State
AB 2575	Health care services: artificial intelligence	LEGISLATIVE PASSAGE COMPLETE
SB 503	Health care artificial intelligence: bias in clinical decision-making and resource allocation	ENROLLED AND PRESENTED TO GOVERNOR
SB 903	Artificial intelligence: mental health and the practice of therapy	LEGISLATIVE PASSAGE COMPLETE
AB 1979	Health care services: artificial intelligence	LEGISLATIVE PASSAGE COMPLETE

Everything KPSGILL has published on this

Derived from this topic’s own entry. A page tagged to this topic appears here without this dossier being edited.

Reform · 2

Analyses · 5

Trackers & guides · 2

Bill dossiers · 4

How progress would be measured

Share of certified record systems capturing generation-time provenance
Share of AI-drafted notes with recoverable model and version identity
Board cases in which provenance was available to the adjudicator

Entity topic.healthcare-ai · all dossiers · event timeline · methodology