California Governor Gavin Newsom has signed into law the Senate Bill 53, the Transparency in Frontier Artificial Intelligence Act. The new measure, known as the California AI bill, requires leading AI companies to disclose details about their most advanced systems, placing the state once again at the forefront of tech regulation. Those in support of the new law highlight that it provides the needed safeguards while still allowing innovation to flourish. Critics warn it could impose compliance burdens that will ripple through the industry nationwide.
California is home to Silicon Valley, today’s nexus of artificial intelligence innovation. What happens in Sacramento rarely stays there. This new AI law could establish a new baseline for AI standards across the United States.
From Brussels Effect to Sacramento Effect
The term Brussels Effect was coined by Anu Bradford, a professor at Columbia Law School, in 2012. She introduced the concept to describe the European Union’s ability to set global regulatory standards through its market power. The European Union’s General Data Protection Regulation was one of its first examples: companies worldwide adopted GDPR’s privacy standards rather than maintain separate compliance systems. California has repeatedly played a similar role inside the U.S.
When the state passed landmark environmental laws in the 1960s, automakers redesigned vehicles nationwide to meet California’s tougher standards. Its 2006 Global Warming Solutions Act became a template for climate legislation across the country. The California Consumer Privacy Act of 2018 pushed data privacy to the top of the national policy agenda, forcing firms to extend protections to all Americans rather than create California-only systems.
The California AI bill could mark the start of a “Sacramento effect” in AI. While Congress and the federal government debate preemption and light-touch rules, California has chosen to act. The new law, authored by state Senator Scott Wiener, is in its second draft after Governor Newsom vetoed its first version (SB 1047) last year after heavy industry lobbying.
Senator Wiener said that “with a technology as transformative as AI, we have a responsibility to support that innovation while putting in place commonsense guardrails to understand and reduce risk. With this law, California is stepping up, once again, as a global leader on both technology innovation and safety.”
The initial reaction from the leading frontier model developers is positive. Chris Lehane, chief global affairs officer at OpenAI posted on LinkedIn that the law “lays out a clearer path to harmonize California’s standards with federal ones. That’s also why we support a single federal approach—potentially through the emerging CAISI framework—rather than a patchwork of state laws.” Anthropic has endorsed SB53 since before its signing stating that the law’s “transparency requirements will have an important impact on frontier AI safety. Without it, labs with increasingly powerful models could face growing incentives to dial back their own safety and disclosure programs in order to compete.”
Inside the California AI Bill
The bill targets frontier AI models, systems with the potential to produce outputs that could cause significant economic or security harms. The text of the law defines frontier models as those trained using computing power greater than 10^26 integer or floating-point operations (including the original training run and any subsequent fine-tuning, reinforcement learning, or other material modifications applied to a preceding foundation model). In summary, models like OpenAI GPT-4 and GPT-5, Google Gemini, Meta Llama series and Anthropic Claude would be in the scope of the law. These definitions should be updated by no later than January 1, 2027, and annually after that.
The act imposes transparency and risk-management duties on frontier developers. Companies must create and publish a frontier AI framework showing how they adopt national and international standards, define catastrophic risk thresholds, apply and review mitigations and if they use third-party evaluations. They must address cybersecurity for model weights, implement governance and incident response processes and assess catastrophic risks, including those resulting from the model itself circumventing oversight mechanisms. The developer must update the framework annually and after material modifications.
Before deploying a new or significantly altered model, developers must release a transparency report with details such as release date, supported languages and modalities, intended uses and restrictions. The report must also include the results of catastrophic risk assessments defined in the frontier AI framework and whether third-party evaluators were involved. The developers are encouraged to publish this information via system or model cards and will be deemed compliant when doing so.
Critical safety incidents must be reported to California’s Office of Emergency Services within 15 days, or within 24 hours if imminent harm is posed. Reports are confidential, exempt from public records and may be transmitted to state or federal authorities. Annual anonymized summaries will inform policymakers without exposing trade secrets.
SB53 protects whistleblowers assessing catastrophic AI risks from retaliation, bans gag policies and allows anonymous reporting. The law creates CalCompute, a public cloud compute cluster to advance safe, ethical and equitable AI. The cluster will be preferably implemented at the University of California and will include the necessary human expertise to operate and maintain the platform, as well as support, train and facilitate its use.
California’s AI law empowers regulators to set reporting rules and impose civil penalties, while maintaining a disclosure-based regime without licensing requirements. Proponents argue the bill increases accountability without stifling innovation. By focusing on transparency, it avoids freezing a rapidly evolving field. Detractors, particularly some industry voices, caution that the reporting requirements could be onerous for startups and tilt the playing field toward established giants that already employ large compliance teams.
AI Regulation in the States and in Washington
California is not alone. State legislatures across the country are introducing AI bills, many aimed at consumer protection, algorithmic bias, or workplace impacts. According to the National Conference of State Legislatures, all 50 states, Puerto Rico, the Virgin Islands and Washington, D.C., have introduced AI legislation this year. Thirty-eight states adopted or enacted around 100.
At the federal level, Senator Ted Cruz and the Senate Commerce Committee have advanced what they describe as a light-touch framework. The proposal emphasizes voluntary guidelines, liability shields and federal preemption of state laws to prevent a patchwork of state regulations. Opponents counter that such an approach risks leaving consumers and workers exposed.
This tug-of-war between strong state action and a restrained federal government echoes earlier technology cycles. During the rise of social media, Washington largely stood back.
Why California Matters in AI
California’s influence extends far beyond its borders. With nearly 40 million residents and a $4.1 trillion economy, it would rank as the world’s fourth-largest economy if it were a country. It is the epicenter of artificial intelligence development. According to the governor’s announcement:
- The state is home to 32 of the world’s 50 top AI companies.
- In 2024, 15.7% of all U.S. AI job postings were in California, nearly double Texas at 8.8% and far ahead of New York at 5.8%.
- More than half of global venture capital funding for AI startups flowed to Bay Area companies.
- Google, Apple and Nvidia, three of the four companies that are valued at over $3 trillion, are based in California. The fourth, Microsoft, is based in Washington state.
California’s AI legislation is not a parochial experiment. For most practical purposes, regulating AI in California means regulating AI in the United States.
What Comes Next
The adoption of the California AI bill raises an important question: Will Congress move quickly to establish a national framework, preempting state laws? Or will Washington once again take a back seat, letting the states experiment?
One possibility is that the California model spreads. Other states may adopt similar transparency requirements, just as they mirrored California’s environmental and privacy standards in the past. If so, the “Sacramento effect” could become a durable feature of America’s AI landscape.
Justice Louis Brandeis once described the states as laboratories of democracy. California has taken up that mantle in artificial intelligence. By enacting the California AI bill, it is testing whether disclosure and accountability can guide a powerful technology without choking off progress.
