On 30 September 2026 California Governor Gavin Newsom announced he had signed a new legislative block on artificial intelligence that, per the official release, strengthens workplace, healthcare and transparency protections. The political tone deliberately contrasted with Washington: “Super intelligence is clearly not coming from the White House,” the announcement said, nodding to the federal naming order signed a day earlier.
On labour, SB 947 and related measures forbid employers from relying only on AI for disciplinary actions or terminations: a human review is required. SB 951 demands more transparency when a mass layoff, relocation or termination is caused by an AI system. AB 1331 and AB 1883 target workplace surveillance, including a ban on surveillance tools in workplace bathrooms.
AB 2392 requires California public higher-education institutions to create AI training and procurement standards for students, faculty and staff. SB 574 keeps lawyers accountable by prohibiting them from fully handing over core legal work—such as drafting briefs or providing legal judgment—to AI systems. The message: the tool assists; the professional licence is not delegated.
In healthcare, AB 1979 and SB 503 aim to ensure doctors and licensed providers can exercise professional judgment when clinical decision tools are used, and that developers take reasonable steps to reduce known or predictable bias. It is a direct response to fears that triage or denial algorithms could operate without a human clinical counterweight.
Transparency and authenticity: SB 1000 and AB 2713 strengthen the California AI Transparency Act, including provenance and a ban on deleting digital watermarks, plus easier access to metadata to tell whether content is AI- or user-generated. SB 1159 addresses transparency and governance; another piece makes clear that public comments at local meetings must come from real people, not bots.
SB 1111 and civil/criminal updates reinforce likeness rights and combat false impersonation, including non-consensual explicit deepfakes. AB 1864, on another risk line, requires gene-synthesis companies to verify customers and genetic material against safety guidelines—a bridge between AI policy and lab biosecurity.
The package builds on earlier September signatures: SB 813 (independent verification organisations) and AB 1405 (state registry of AI auditors), plus an executive order accelerating third-party audits and a two-month expert group to propose reinforcements, including third-party-written safety plans and a “kill switch” for frontier models.
Lorena Gonzalez, president of the California Federation of Labor Unions, hailed SB 947, SB 951, AB 1331 and AB 1883 as the start of nation-leading workplace guardrails. For firms with California headquarters or payroll, the compliance calendar thickens: HR policies, people-analytics vendors and monitoring tools must be audited before each statute’s effective date.
The executive order that “declares the obvious”—calling the technology artificial intelligence rather than “super intelligence”—is symbolic but signals federal–state conflict. While the White House pushes SI in federal language, Sacramento codifies AI and adds labour teeth that the voluntary frontier-lab audit pact does not contemplate.
Compared with Europe’s AI Act, California’s approach is more granular on employment, health and deepfakes, and less centred on systemic-risk GPAI categories. That can create mosaic compliance for multinationals: the same HR agent may be lawful in one state and banned in California if it fires without a human in the loop.
Enforcement remains to be proven: agency rulemaking timelines and employer litigation will test whether the bans are operational or declarative. Neural Pulse does not assume automatic enforcement; it marks 2026–2027 implementation as the real test.
Taken together, 30 September leaves California as the most aggressive U.S. AI workplace regulatory lab, just as the federal debate oscillates between “morally binding” pacts and unfinished bills. Anyone deploying decision agents over people in the state should review the package before optimising the model.
