Recommended Free Tools
Anthropic endorsed California Senate Bill 53 on September 8, 2025, while the bill was still moving through the Legislature. The company backed its transparency-focused approach to frontier-AI risks but continued to favor federal rules over a patchwork of state laws. Governor Gavin Newsom signed the bill on September 29, 2025; it took effect January 1, 2026, as the Transparency in Frontier Artificial Intelligence Act.
What Anthropic endorsed
Anthropic’s announcement was support for a particular bill, not for California AI regulation in general. It said frontier-AI developers should disclose how they assess and manage catastrophic risks, while retaining flexibility to choose their technical safety methods. The company described this as a “trust but verify” approach and said a federal framework would be preferable to state-by-state rules. Anthropic’s September 8 announcement explains its stated rationale.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
California School Law: Third Edition | $43.16 | Buy on Amazon |
| 2 |
|
California Landlord's Law Book, The: Rights & Responsibilities | $34.88 | Buy on Amazon |
| 3 |
|
Merriam-Webster's Dictionary of Law, Newest Edition, Trade Paperback | $10.99 | Buy on Amazon |
| 4 |
|
California Statutes - Evidence Code 2026 | $19.99 | Buy on Amazon |
The endorsement applied to the bill as it stood on September 8. The Legislature amended SB 53 before enactment, so the enrolled statute—not the announcement or an earlier draft—is the operative law. The enacted text sets out the legal duties.
What the law requires—and who it targets
SB 53 is more accurately described as a frontier-AI transparency and catastrophic-risk law than as a general AI safety code. Its obligations apply to covered large frontier developers under statutory definitions; using an AI product or providing ordinary software does not, by itself, make a company subject to the law’s principal developer requirements. Coverage depends on the statute’s definitions, thresholds, activities, and applicable exemptions.
#1 Best Overall
For covered developers, the law establishes requirements involving safety and security protocols, public information about catastrophic-risk assessment and management, reporting of specified critical safety incidents, and protections for covered employees who raise concerns. It also provides for state oversight, preempts certain local rules relating to frontier developers and catastrophic-risk management, and establishes the CalCompute initiative to expand access to computing resources for AI research and innovation. The details and exceptions are in the statute.
Catastrophic risk is a narrow legal category
The statute defines catastrophic risk using very high-impact thresholds, including death or serious injury to more than 50 people, or more than $1 billion in property damage or loss arising from a single incident involving a frontier model. Those thresholds define a particular catastrophic-risk framework; they do not make other harms legally irrelevant. SB 53 is not a comprehensive law for discrimination, misinformation, copyright disputes, consumer fraud, deepfakes, or every other form of AI misuse.
Disclosure is not proof of effectiveness
Publishing a safety framework gives regulators and the public information about a developer’s stated process. It does not, by itself, demonstrate that safeguards work or that a model cannot cause harm. The practical value of the disclosure duties depends on the quality and specificity of what companies publish, and on implementation and enforcement.
Why this was different from the SB 1047 debate
Anthropic had not supported California’s earlier SB 1047 in its original form. The company’s stated objection was to requirements it regarded as overly prescriptive or poorly targeted, not to safety regulation as a whole. Its SB 53 endorsement reflected a distinction between requiring disclosure about risk management and prescribing particular technical methods. Contemporary reporting on Anthropic’s position in the earlier debate is available from Axios.
| Issue | SB 1047 debate | SB 53 |
|---|---|---|
| Anthropic’s position | Did not support the earlier bill in its original form. | Publicly endorsed the bill on September 8, 2025. |
| Regulatory emphasis | Anthropic objected to provisions it viewed as overly prescriptive or poorly targeted. | Emphasizes disclosure, incident reporting, and developer safety processes while leaving technical methods more flexible. |
| Policy scope | Earlier proposal addressed AI safety and liability concerns through a different framework. | Focuses on covered frontier developers, catastrophic-risk management, transparency, and related duties. |
| Level of government | Anthropic’s preference for federal rules remained. | Anthropic accepted state action while continuing to argue for a federal framework. |
This contrast describes Anthropic’s stated view, not a neutral finding that one bill was necessarily more effective. The central policy choice is whether government should require disclosure and accountability while leaving methods to developers, or prescribe more specific safety measures.
Why Anthropic supported disclosure while preferring federal rules
Anthropic’s case was that frontier-model capabilities and evaluation methods change quickly, making fixed technical mandates difficult to keep current. It argued that requiring developers to explain their risk-management practices offers oversight without locking companies into one method. It also presented California action as useful while federal agreement remained absent, despite its preference for national rules. Its later account of how it would approach the statute appears in its SB 53 compliance framework.
That rationale has a trade-off. Flexibility can let developers adapt practices as systems change, but broad or carefully worded disclosures may be hard for outsiders to evaluate. A public protocol is evidence of what a company says it does; it is not independent validation that the process prevents harm.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Business opposition and changes before enactment
Anthropic’s support did not mean the technology and business sectors agreed. An Assembly committee analysis recorded opposition or “oppose unless amended” positions from groups including the Silicon Valley Leadership Group, Chamber of Progress, California Chamber of Commerce, CCIA, and TechNet. Their positions reflected concerns about the bill’s burdens, clarity, or effects on innovation. The committee analysis also records changes made during the legislative process.
Among the changes identified in that analysis were narrowing some critical-incident categories to incidents involving actual harm, adjusting language concerning false or misleading statements, reducing the maximum civil penalty from $10 million to $1 million, removing contractors from whistleblower protections, narrowing public disclosure of certain internal-use model risk assessments, and preempting certain local frontier-AI regulations. These changes are one reason to distinguish the September 8 bill Anthropic endorsed from the final statute.
Support also came from civil-society, labor, and AI-safety organizations. California legislative materials described the bill against a backdrop of voluntary frontier-AI safety commitments by major developers, including Anthropic and other large companies. Senator Scott Wiener’s account of the bill’s development discusses that context. SB 53’s significance is not simply that companies may have safety practices, but that the statute makes specified processes, disclosures, and reporting duties mandatory for those it covers.
Enactment and Anthropic’s compliance response
Newsom signed SB 53 on September 29, 2025, giving it the formal name Transparency in Frontier Artificial Intelligence Act. The law took effect January 1, 2026. The governor’s signing announcement describes the enactment; the enrolled law controls the legal requirements.
On December 19, 2025, Anthropic published a compliance framework ahead of the effective date. It connected its approach to its existing Responsible Scaling Policy and described how it would assess and manage catastrophic risks. Anthropic said the law balanced safety practices, incident reporting, and whistleblower protections with flexibility for developers and less burden on smaller companies. That publication is the company’s account of its approach, not an independent audit or proof of effectiveness.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesWhat the endorsement does—and does not—show
Anthropic’s endorsement gave the bill support from a company directly affected by its rules, but that interest matters when weighing the signal. A framework compatible with existing developer practices can reduce uncertainty for established frontier companies and may formalize commitments they already say they follow. That does not establish that Anthropic supported the bill for self-interested reasons, nor that its endorsement caused the bill to pass.
The law’s promise is greater visibility and accountability around a narrow set of severe risks. Its limits are equally important: disclosure may not reveal enough to assess safeguards, the catastrophic-risk thresholds exclude many ordinary AI harms, and the law’s real effect depends on implementation, enforcement, and agency capacity. Anthropic endorsed a specific transparency model, not a guarantee that California has solved AI safety or a retreat from its preference for federal regulation.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
