What it is
SB 53 — formally the Transparency in Frontier Artificial Intelligence Act (TFAIA) — is a California law requiring large AI developers to disclose how they manage safety risk in their most powerful models. Governor Gavin Newsom signed it on September 29, 2025, and its core provisions took effect January 1, 2026. It is widely described as the first state or national law in the United States specifically targeting “frontier” AI safety.
The law applies to companies that train models using more than 10²⁶ floating-point operations (FLOPs) of compute — a threshold meant to capture only the handful of firms building the most capable systems, including OpenAI, Anthropic, Google DeepMind, and Meta. The heaviest obligations fall on “large frontier developers,” defined additionally by annual revenue above $500 million. Those companies must publish a documented framework describing how they incorporate safety standards into model development, report “critical safety incidents” to California’s Office of Emergency Services (within 15 days generally, or 24 hours if there is imminent risk of death or serious injury), and extend whistleblower protections to employees who raise safety concerns internally. Violations can carry penalties of up to $1 million per incident.
SB 53 is the direct successor to SB 1047, a more sweeping frontier-AI liability bill by the same author, State Senator Scott Wiener, which Newsom vetoed in September 2024. After the veto, Newsom convened the Joint California Policy Working Group on AI Frontier Models — co-led by Stanford’s Fei-Fei Li, the Carnegie Endowment’s Mariano-Florentino Cuéllar, and UC Berkeley’s Jennifer Tour Chayes — to determine what state policy could realistically accomplish. Its June 2025 final report recommended narrower, transparency-and-incident-reporting-focused interventions rather than the liability regime SB 1047 had proposed; Wiener rewrote the bill along those lines, and the revised version became SB 53.
Why it matters for AI governance and narratives
SB 53 sits at the center of a live framing contest over how — and at what level of government — frontier AI risk should be regulated. Its passage is frequently cited as evidence that transparency and incident-reporting obligations, rather than SB 1047-style liability rules, represent the achievable ceiling for state-level AI safety legislation in the current political environment, particularly given industry lobbying against SB 1047 and a federal posture that has at times sought to preempt state AI rules altogether. Newsom has explicitly framed the law as a national model, arguing it can shape policy beyond California “in the absence of a comprehensive federal framework” — a claim that itself is a move in the contest over whether AI governance authority defaults to states, Washington, or industry self-regulation.
The law is also a proxy fight over what counts as adequate self-governance. Frontier labs’ voluntary safety frameworks and benchmark commitments — the kind of disclosures SB 53 partially codifies into binding obligation — become harder to treat as purely voluntary once a regulator can request the same documentation under law. That is the significance of AI labs, including OpenAI, publicly lobbying to strengthen SB 53 with additional frontier-model monitoring and cybersecurity-incident provisions even as they resist broader liability regimes elsewhere: it is a bid to shape the specific content of the state’s disclosure requirements rather than to avoid disclosure obligations altogether.
Key facts and dates
- September 29, 2024: Newsom vetoes SB 1047, Wiener’s earlier and more expansive frontier-AI safety bill.
- 2025: Newsom’s Joint California Policy Working Group on AI Frontier Models, led by Fei-Fei Li, Mariano-Florentino Cuéllar, and Jennifer Tour Chayes, studies what state AI policy can practically address.
- June 2025: The Working Group publishes its final report, recommending targeted transparency and incident-reporting measures.
- September 29, 2025: Newsom signs SB 53 into law.
- January 1, 2026: Core provisions take effect, including safety-framework publication and incident-reporting duties.
- Threshold for coverage: Models trained with more than 10²⁶ cumulative FLOPs (including fine-tuning); heightened obligations apply to developers with more than $500 million in annual revenue.
- Enforcement: Reports of critical safety incidents go to the California Office of Emergency Services; penalties can reach $1 million per violation.
Sources agree on these core facts; some secondary summaries vary slightly on subsidiary details (e.g., exact reporting thresholds for smaller developers), so readers seeking precise compliance obligations should consult the bill text or a primary legal analysis directly.
Where to learn more
- SB 53 full bill text — California Legislative Information — the primary legal text of the law.
- Governor Newsom Signs Senator Wiener’s Landmark AI Law — Senator Scott Wiener’s office — the author’s official statement on signing and intent.
- California’s SB 53: The First Frontier AI Law, Explained — Future of Privacy Forum — nonprofit policy-research explainer of provisions and scope.
- Governing Frontier AI: California’s SB 53 — Lawfare — analytical piece situating SB 53 within the broader AI-governance landscape and its relationship to SB 1047.