A seguirPróximo guia
California ADMT Rules for Automated Decision-Making
Sociedade
GUIA DA SOCIEDADE
Califórnia SB 53 e Frontier AI Transparency
California SB 53, the Transparency in Frontier Artificial Intelligence Act signed in September 2025, applies to developers of the most computationally intensive AI models.
Nesta página4 minutos de leitura
Visão geral
It requires them to publish safety frameworks and transparency reports, report critical safety incidents to the state, and protects whistleblowers who raise catastrophic risk concerns. It matters because it is one of the first US laws aimed specifically at frontier model developers, and it replaced the liability-focused approach of the vetoed SB 1047 with required disclosure.
Mergulho profundo
SB 53 defines a frontier model as a foundation model trained with more than 10^26 integer or floating-point operations, including compute from later fine-tuning or modification. Any developer that has trained such a model is a frontier developer and must publish a transparency report when it deploys a new or substantially modified frontier model. A large frontier developer, one with more than $500 million in annual revenue, must also publish and annually review a frontier AI framework, include summaries of catastrophic risk assessments in its transparency reports, and periodically send the Office of Emergency Services (OES) summaries of assessments of catastrophic risk from internal use of its models. Catastrophic risk means a foreseeable and material risk that a model contributes to the death or serious injury of more than 50 people, or more than $1 billion in damage, in a single incident. The risk must arise through expert-level help with chemical, biological, radiological or nuclear weapons; a cyberattack or serious crime carried out without meaningful human oversight; or the model evading its developer's control. Critical safety incidents include unauthorized access to model weights, harm from a materialized catastrophic risk, and loss of control. Developers may not make materially false or misleading statements about catastrophic risk. The Attorney General enforces the law with civil penalties of up to $1 million per violation. SB 53 also creates a consortium to design CalCompute, a public computing cluster. The law followed SB 1047, also by Senator Scott Wiener, which would have required pre-deployment safety testing, shutdown capability and liability for critical harms. Governor Newsom vetoed it in September 2024 and convened an expert group whose June 2025 report favored transparency with verification. SB 53 does not require licensing, pre-approval or a kill switch.
Impacto Estratégico
Risco e segurança
Os danos catastróficos e diários da IA dependem de quem entende os riscos e de quem pode agir.
Decisões mais claras
A literacia pública e profissional determina se uma política de segurança forte é politicamente possível.
Cortando o hype
Explicações claras reduzem a captura por exageros, relações públicas de laboratório e teatro de ética vaga.
The Future of California SB 53 and Frontier AI Transparency
Most obligations took effect on January 1, 2026, so there is little enforcement history yet. Open questions include how OES will handle incident reports and how actively the Attorney General will pursue penalties. The law directs state officials to review its definitions and thresholds over time, because a fixed compute threshold may become outdated as training becomes more efficient. The biggest uncertainty is federal policy: efforts in Congress and from the White House to preempt or challenge state AI laws have left the relationship unsettled. Other states, including New York, have pursued similar frontier-model transparency legislation.
Implementação no mundo real
A large frontier developer with more than $500 million in annual revenue must publish a frontier AI framework on its website. The framework explains how the company assesses and mitigates catastrophic risks, including whether it uses third-party evaluators.
When a developer releases a new frontier model, it must publish a transparency report covering the model's intended uses and restrictions. Large developers must also summarize their catastrophic risk assessments in that report.
A developer that discovers its unreleased model weights were stolen must report the critical safety incident to California's Office of Emergency Services within 15 days. If the incident poses an imminent risk of death or serious injury, the deadline is 24 hours.
An employee who believes their company is misrepresenting catastrophic risk can report it to the Attorney General or other authorities and is protected from retaliation. Large developers must also provide an anonymous internal reporting channel.
Riscos e guarda-corpos
Tratar o risco existencial como ficção científica enquanto aumenta a capacidade.
Confundir segurança do produto de superfície com alinhamento sob alta autonomia.
Deixando o público não-inglês e não especializado com apenas fontes de baixa qualidade.
Roteiro de implementação
Separe os riscos de danos ao produto, uso indevido e perda de controle/desalinhamento.
Pergunte quais evidências mudariam sua visão sobre prazos e gravidade.
Prefira fontes primárias e avaliações concretas em vez de afirmações de marketing.
Identifique um caminho de ação: carreira, política, financiamento ou habilidades – não apenas conscientização.
Continue explorando
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the California SB 53 and Frontier AI Transparency quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Perguntas frequentes
What is California SB 53 and Frontier AI Transparency?
California SB 53, the Transparency in Frontier Artificial Intelligence Act signed in September 2025, applies to developers of the most computationally intensive AI models. It requires them to publish safety frameworks and transparency reports, report critical safety incidents to the state, and protects whistleblowers who raise catastrophic risk concerns. It matters because it is one of the first US laws aimed specifically at frontier model developers, and it replaced the liability-focused approach of the vetoed SB 1047 with required disclosure.
What training compute threshold defines a frontier model under SB 53?
SB 53 defines a frontier model as one trained with more than 10^26 integer or floating-point operations, counting compute used in later fine-tuning or modification.
What annual revenue makes a frontier developer a large frontier developer?
Developers with more than $500 million in annual revenue count as large frontier developers. They must publish a frontier AI framework and meet additional reporting duties.
What is the standard deadline for reporting a critical safety incident to the state?
Developers generally must report critical safety incidents to the Office of Emergency Services within 15 days. A shorter deadline applies when there is imminent danger.
How quickly must an incident be reported if it poses an imminent risk of death or serious injury?
If an incident poses an imminent risk of death or serious injury, the developer must notify an appropriate authority within 24 hours.
Which of these does SB 53 NOT require?
SB 53 relies on disclosure, incident reporting and whistleblower protection. It does not require licensing, pre-approval or a kill switch. Shutdown capability was part of the vetoed SB 1047.
Continue aprendendo
Guias relacionados
Mais guias escolhidos para este tópico