California's final-day AI package can be read as a list of bills. It is more useful read as a set of rules about one question: at which decision points must a human still decide? The answer the state gave on September 30, 2026, is specific, and so are the two places where it declined to go further. The health-care measures are examined in detail in this site's lead analysis; this piece maps the whole boundary.
The decision points
| Domain | Measure | The rule, as described by the Governor's office or sponsor | Status |
|---|---|---|---|
| Clinical care | AB 1979 | Licensed professionals keep their professional judgment when clinical decision tools inform care; consumer health chatbots are covered by medical-confidentiality law | Signed September 30, 2026 |
| Clinical tools | SB 503 | Developers and deployers must make reasonable efforts to identify foreseeable bias | Signed September 30, 2026 |
| Discipline and termination | SB 947 | Employers may not rely solely on AI | Signed September 30, 2026 |
| Layoffs | SB 951 | Notice required when AI causes a mass layoff, relocation or termination | Signed September 30, 2026 |
| Workplace surveillance | AB 1331; AB 1883 | No surveillance tools in workplace bathrooms; limits on surveillance tools, including emotional-state prediction and neural data | Signed September 30, 2026 |
| Verification | SB 813; AB 1405 | Voluntary certification of independent verification organizations; a registry of AI auditors | Signed September 9, 2026 |
| Override protection | AB 2575 | Would have protected clinicians who override AI and limited developers' ability to blame the clinician | Vetoed September 30, 2026 |
| Mental-health workforce | SB 903 | Would have barred replacing mental-health professionals with AI | Vetoed September 30, 2026 |
Sources: Governor's AI release; September 30 legislative update; CalMatters.
The September 30 package also covered AI provenance and transparency, digital replicas, procurement standards for public universities, screening in gene-synthesis systems, and the use of AI in the legal system. Those measures sit outside health care and are listed in the site's final legislative ledger.
The pattern
AnalysisThe signed measures require process: notice, disclosure, documentation, human review, an available auditor. The vetoed measures would have reallocated risk or constrained staffing decisions: one by protecting the clinician who overrides the machine and limiting the developer's defense, the other by barring the substitution of AI for mental-health workers. This describes what was enacted and what was not. The Governor's reasons for the two vetoes are stated in messages that have not yet been transcribed, and this site does not attribute a motive before they have been read.
What it means for a physician
From January 1, 2027, California's rules presume that a human decides at the bedside and in the human-resources office, and that someone has looked for bias in the tool. They do not protect the human who decides against the tool. For a clinician employed by an organization that owns the algorithm, that is the gap that matters, and it is the subject of the site's lead analysis of AB 1979 and SB 503.