Summary: California’s SB 53 took effect on January 1, 2026, requiring transparency reports before or alongside deployment of a new or materially modified frontier model trained with more than 10^26 integer or floating-point operations. Developers above $500 million in prior-year affiliated revenue must also publish and follow annual safety frameworks; critical safety incidents carry 15-day reports to CalOES, or 24-hour disclosure to appropriate authorities for imminent death or serious physical injury risks. Violations can bring civil penalties up to $1 million per violation, enforceable by the California Attorney General.
New Year’s Day 2026 marked the start of California’s SB 53 — the Transparency in Frontier Artificial Intelligence Act. Sponsored by Senator Scott Wiener and signed by Governor Gavin Newsom on September 29, 2025, it stands as the first US state statute written specifically for frontier AI models, and as the recalibrated route after SB 1047 died to a veto in 2024: no kill switch and no pre-deployment licensing this time. The bargain is simpler — write your risk assessments down, publish them, and pay if you don’t.
The structural stakes explain the noise. Most frontier labs are headquartered or operate substantially inside California, which places every model shipped after the effective date immediately under the new disclosure duties. Paired with OpenAI’s Rest-of-World terms, which also switched on January 1 (see our opening outlook), the law sets 2026’s tone: the rules are tightening.
Who Is Covered: 10^26 FLOPs and the $500M Line
The statute builds its scope from two nested definitions. A “frontier model” is a foundation model trained with more than 10^26 integer or floating-point operations — and the tally sweeps in subsequent fine-tuning, reinforcement learning, and material modifications, not merely the original training run. Whoever initiates training of such a model is a “frontier developer.”
The heavier obligations land only on “large frontier developers” — those whose combined annual gross revenues with affiliates topped $500 million in the prior calendar year. California’s Department of Technology must revisit these definitions every year, with a first round of recommended adjustments due by January 1, 2027. The thresholds were built to drift with reality rather than freeze at enactment.
Two New Documents: Transparency Reports and Safety Frameworks
Any frontier developer deploying a new or materially modified frontier model must publish a transparency report before or alongside the deployment, covering at minimum the developer’s website, a mechanism for communicating with users, the release date, and intended uses. An existing system card or model card that already covers those items clears the bar. Redactions are permitted for trade secrets, cybersecurity, or national security — but each one must be described and justified inside the report, and unredacted versions are to be retained for five years.
Large developers layer on more: summaries of catastrophic-risk assessments and their results, plus the degree of third-party evaluator involvement. They must also draft, genuinely follow, and publish on their websites a Frontier AI Safety and Security Framework, reviewed and refreshed at least annually — spanning standards adoption, catastrophic-risk thresholds and assessment, pre- and post-deployment review, cybersecurity for unreleased model weights, incident response, and internal governance.
The law’s notion of catastrophic risk is anything but vague: a foreseeable, material risk of contributing to the death or serious injury of 50 or more people, or more than $1 billion in property damage from a single incident — arriving through CBRN weapons assistance, unsupervised cyberattacks or serious crimes, or evasion of developer and user control.
The Incident Clock and Whistleblower Protections
Reporting duties run on two clocks. Upon discovering a critical safety incident, developers owe CalOES — the California Office of Emergency Services — a report within 15 days. Where the incident poses an imminent risk of death or serious physical injury, disclosure to an appropriate law enforcement or public safety authority is due within 24 hours. Large developers additionally transmit summaries of their internal-use catastrophic-risk assessments to CalOES every three months, or on another reasonable cadence.
Whistleblower protection ranks among the statute’s sharpest teeth. Developers may not adopt policies or contracts that prevent, or retaliate for, an employee’s disclosure — to the California Attorney General, federal authorities, a supervisor, or colleagues carrying out an investigation — of information reasonably believed to show a specific, substantial public safety danger or an SB 53 violation. Large developers must operate an anonymous internal reporting channel with monthly status updates and post annual notices of employee rights.
Penalties, Enforcement, and the 2027 Wave
Breaching the framework-publication, transparency-report, or critical-incident reporting duties draws civil penalties of up to $1 million per violation, enforceable solely through civil actions brought by the California Attorney General. The fine print carries the punchline: failing to follow your own published safety framework is itself a violation — the moment the document goes live, it hardens into an enforceable commitment.
January 1, 2027 is the next waypoint: CalOES begins releasing anonymized annual incident summaries, and the Department of Technology delivers its definition updates; the law also stands up the CalCompute public compute consortium. One clarification worth repeating: SB 942, the California AI Transparency Act governing provenance disclosure and watermarking of AI-generated content, was pushed by AB 853 to August 2, 2026. What woke up on New Year’s Day was SB 53, not SB 942.
What It Means for Dev and Product Teams
Three practical moves. First, scope yourself honestly — keep training-compute records and aggregate affiliate revenue, because the $500 million line is easier to cross than most teams assume. Second, align your model card pipeline with the transparency-report requirements and paper every redaction; both what you disclosed and what you omitted can surface in litigation. Third, wire the 15-day and 24-hour clocks into your incident response plan and stand up an anonymous whistleblower channel. None of this belongs to the legal department alone — model, safety, and engineering teams all hold pieces of the job.