SB 53 SIGNED IN CALIFORNIA
AN UNPRECEDENTED INCURSION INTO COMMERCE, FREEDOM, SPEECH,
AND AI DEVELOPMENT
TO CREATE :
AN UNENDING OPEN-ENDED COMPLIANCE EXERCISE AND LAWYERS MADE RICH
AND/OR
AN ALMOST UNENDING OPEN-ENDED LITIGATION CRUSH, TAKING RESOURCES OF MONEY AND TIME
AWAY FROM DEVELOPMENT OF THE MOST PROMISING TECHNOLOGY SINCE THE INTERNET
DRAW YOUR OWN CONCLUSIONS
SUMMARY
TODAY, On September 30, 2025, California Governor Gavin Newsom signed SB 53 into law,
establishing the nation’s first comprehensive AI “safety” transparency requirements for large-scale models.
This isn’t just another state-level tweak—it’s a seismic shift that directly targets Silicon Valley’s AI giants (home to 32 of the world’s 50 largest AI firms),
forcing companies with $500M+ in annual revenue to publicly disclose their “safety” protocols, risk assessments, and mitigation strategies.
Dubbed a “first-of-its-kind” measure, it “balances innovation with accountability”,
but it’s already sparking fierce debate on enforcement,
preemption battles with federal regulators,
and the flood of compliance work and litigation crush heading to Law.
Lawyers sometimes like it when bad things happen.
Not this lawyer.
ABBREVIATED TEXT
“Trust but verify” [extremely NOXIOUS to treat our leading AI companies as Soviet Russia!]
Abbreviated Text of California Senate Bill 53 (2025): Transparency in Frontier Artificial Intelligence Act (TFAIA)
Enacted September 29, 2025 Author: Senator Scott Wiener (D-San Francisco) Key Purpose: Establishes transparency requirements for developers of “frontier AI models” (large-scale generative AI systems with significant training compute, e.g., >10^26 FLOPs) to mitigate catastrophic risks like bioweapons, cyberattacks, or loss of control, while protecting innovation. Applies to “frontier developers” with ≥$500M annual global revenue. Phased rollout starts January 1, 2027.
§ 22757.10 – Legislative Findings and Declarations
AI frontier models pose risks to public health/safety (e.g., CBRN weapons, deception).
California hosts 32 of top 50 global AI firms; needs “trust but verify” governance.
Builds on voluntary industry commitments; preempts local regs on catastrophic risk management post-Jan 1, 2025.
§ 22757.11 – Definitions
Frontier Developer: Entity with ≥$500M revenue developing frontier models.
Frontier Model: Generative AI trained on massive compute, capable of broad risks.
Critical Safety Incident: Event where model enables substantial harm (e.g., autonomous cyberattack, CBRN aid without oversight).
Covered Employee: AI lab worker with access to risk info.
Chapter 25.1 (Business & Professions Code): Transparency & Reporting
Annual Public Disclosures (§ 22757.12): Developers must post on website:
Safety protocols (testing, red-teaming, mitigations for bias/deepfakes/catastrophic risks).
Risk assessments (e.g., CBRN, deception, misalignment).
High-level summaries for <$500M firms; detailed for larger.
Incident Reporting (§ 22757.13): Report critical incidents to CA Office of Emergency Services (Cal OES) within 72 hours; exempt from Public Records Act.
Includes crimes/deception not covered by EU AI Act.
Enforcement (§ 22757.14): Civil penalties up to $10,000/violation by Attorney General; no private right of action.
§ 11546.8 (Government Code): Dept. of Technology Oversight
Annual recommendations to Legislature on updates, based on multistakeholder input, tech advances, and global standards (e.g., EU AI Act alignment).
Establishes “CalCompute”: Public cloud cluster for democratized AI access/research.
Chapter 5.1 (Labor Code): Whistleblower Protections
Anti-Retaliation (§ 1107 et seq.): Prohibits policies/contracts gagging disclosures of “substantial danger” risks.
Employees may report to AG, federal agencies, supervisors, or peers.
“Reasonable cause” belief required; covers catastrophic risks or TFAIA violations.
Exempt reports from Public Records Act; remedies include reinstatement, backpay.
Preemption & Severability
Overrides local laws on frontier developer risk management.
If any provision invalid, remainder stands.
Effective Date: January 1, 2026 (phased compliance by 2027). This ~5-page law (full text ~20 pages with amendments) codifies expert recommendations from Newsom’s 2025 AI working group, differing from vetoed SB 1047 (2024) by focusing on disclosure over mandatory testing. For full text, visit leginfo.legislature.ca.gov
