AI Regulation in California: The Daiki SB-53 Recipe for the Transparency in Frontier Artificial Intelligence Act
By Editor
Stanford, CA, December 1, 2025—Daiki, the AI and quantum governance company co-founded by Mauritz Kop, has published The Daiki SB-53 Recipe: a practical compliance roadmap for California's Senate Bill 53, the Transparency in Frontier Artificial Intelligence Act (TFAIA). The piece opens with the thesis that frames Daiki's whole approach to artificial intelligence regulation: "Global AI governance is shifting from voluntary principles to enforceable rules."
Six gates to frontier-AI compliance: the SB-53 recipe, visualized.
What SB-53 actually regulates
Signed into law by Governor Newsom on September 29, 2025, and authored by Senator Scott Wiener, SB-53 is the first U.S. law to target frontier AI models explicitly through a compute threshold. Its scope is defined by that threshold: foundation models trained using more than 1026 floating-point operations, including compute spent on substantial fine-tuning or reinforcement learning. A "large frontier developer"—one whose annual group revenue exceeds USD 500 million—carries the heaviest obligations. The law's pivot point is catastrophic risk: a foreseeable, material risk of death or serious injury to more than 50 people, or more than USD 1 billion in property damage, from a single incident. Most requirements take effect on January 1, 2026.
The bill did not emerge in a vacuum. After vetoing the more interventionist SB-1047 in 2024 as overly burdensome, Governor Newsom convened the Joint California Policy Working Group on AI Frontier Models, whose California Report on Frontier AI Policy recommended evidence-generating transparency and incident reporting over direct prohibitions. SB-53 largely implements those recommendations in a narrower, procedural form—no ex ante licensing, no kill switches, but disclosure obligations a regulator can actually enforce.
The obligations: frameworks, reports, incidents, whistleblowers
Large frontier developers must publish a Frontier AI Framework describing how they integrate national and international standards—the NIST AI Risk Management Framework and ISO/IEC 42001—define and assess catastrophic-risk thresholds, mitigate risks before deployment, secure model weights, and govern themselves internally, with annual review and republication of material changes. Deployment of a frontier model triggers a transparency report covering intended uses, restrictions, and risk-assessment summaries. Critical safety incidents must be reported to California's Office of Emergency Services within fifteen days of discovery—within twenty-four hours where there is imminent risk of death or serious injury. Covered employees receive whistleblower protections, including an anonymous internal disclosure channel. The California Attorney General can impose civil penalties of up to USD 1 million per violation.
The Daiki SB-53 Recipe, published at dai.ki on December 1, 2025.
The Daiki recipe: six steps to SB-53 readiness
The heart of the piece is a six-step implementation roadmap. First, determine applicability: inventory foundation models, estimate training compute against the 1026-operation threshold, and assess revenue exposure. Second, build the AI governance baseline on ISO/IEC 42001 and the NIST AI RMF—the same standards-centric logic that runs through the standards-first governance work Kop and colleagues published in Science for quantum technology. Third, design the Frontier AI Framework as an operational document rather than a brochure. Fourth, operationalize transparency reports with repeatable model-card workflows and redaction protocols. Fifth, build the incident and whistleblower workflows—detection, triage, assessment, reporting. Sixth, harmonize with other regimes, integrating SB-53 compliance with the EU AI Act, ISO/IEC 42001, and emerging state laws, so the organization runs one governance system instead of a stack of silos.
Daiki is building the recipe into its platform: a frontier-model registry with compute-threshold tracking, framework templates aligned with SB-53 and the underlying standards, transparency-report generation designed for SB-53, incident-workflow orchestration, and whistleblower attestation—integrated with its existing EU AI Act and quantum-governance solutions.
From compliance burden to governance asset
The register of the piece is deliberately constructive. As the recipe puts it: "SB-53 is best understood not as an isolated burden but as an opportunity to institutionalise frontier-grade AI safety practices." That reading has a history in Kop's work: the idea that impact assessment can be a design instrument rather than paperwork goes back to the AI Impact Assessment he presented at Apple, Facebook, and Stanford in 2019, long before frontier-model law existed.
For boards and general counsel, the practical message is short: AI governance has stopped being a voluntary virtue and started being a regulated function with named penalties. Organizations that build one coherent, standards-based governance system now—rather than a reactive checklist per statute—will find that the era of enforcement rewards exactly the discipline the era of principles merely recommended.
Last updated: June 5, 2026.