SB 53

in English: California SB 53 (Transparency in Frontier Artificial Intelligence Act)

advancedspreading

A California law that requires the largest AI developer companies to publicly report on their security practices and report serious incidents.

In plain terms

SB 53 is a California state law that applies to companies developing the most powerful, so-called frontier models. It does not dictate how to build artificial intelligence. It requires them to show how they look after safety.

It is like a restaurant kitchen where authorities do not prescribe recipes, but hygiene regulations must be posted, and any illness must be reported. The company decides what security system to operate, but must publicly commit to it and cannot make false claims about it.

It matters because most of the world's leading AI companies operate in California, so a state law effectively has global impact. It differs from the EU AI Act in being much narrower: it targets only the largest developers and catastrophic risks, not everyday AI applications.

How it works

The law first defines who it applies to: developers who train models using extremely large amounts of computing power. The strictest obligations fall on those with annual revenue above USD 500 million, known as large developers.
Major developers must publish a safety framework on their website. It describes how they assess and mitigate catastrophic risks, and they must review it at least annually.
When a new frontier model is released, every covered developer must publish a transparency report on what the model is for. Large developers must also summarize the risk assessment they carried out on it.
Serious security incidents must be reported to the California Office of Emergency Services within 15 days of their discovery. If there is an immediate threat to life or risk of serious injury, the relevant authority must be notified within 24 hours.
The law protects employees whose job is to assess or manage safety risks and who, in good faith, report a serious danger. Major developers must also operate an anonymous internal reporting channel. The state attorney general can impose civil fines of up to $1 million per violation in some cases.

An everyday example

A new, high-performance AI model appears, and you read in the news that the developer company also published a safety report with it. If the company falls under California regulation, it does so partly due to SB 53. You can then read on its website what risks were examined before release.

Who is affected

developerIf you work on assessing or managing security risks at a company developing top-tier models, the law protects you when you report a serious danger in good faith. → Code

entrepreneur, executiveThe public safety frameworks of the major AI companies you choose as suppliers help you judge how reliable the model is that you build your business on.

lawyer, citizenAn important model for regulating AI with transparency obligations, and it is well comparable with the EU AI Act. → Law

everyoneWe use models developed in California in Hungary, so the transparency rules there are indirectly useful to us as well.

A common misconception

MythSB 53 bans the release of dangerous AI models and requires prior regulatory approval for them.

In realityThe law does not approve or ban models: it demands transparency, incident reporting, and adherence to the company's own commitments. The company decides on release itself but cannot hide how it manages risks.

Limits and risks

It applies only to the largest developers; smaller companies and firms that merely use AI largely fall outside its scope.
It focuses on catastrophic risks (more than 50 deaths or serious injuries, or damage exceeding one billion dollars), and does not address everyday harms such as discrimination or data privacy.
Companies largely write their own safety frameworks, so the law holds them to their commitments, not to how strict those commitments are.
A state-level law that may be overridden or restricted by later US federal regulation.

Similar, but not the same

SB 1047 – The proposal regarded as its predecessor, vetoed in 2024, was stricter; for example, it would have required a shutdown mechanism and liability rules, whereas SB 53 is limited to transparency.
EU AI-rendelet (AI Act) – The EU regulation governs various AI systems according to risk levels, from prohibited practices to high-risk applications and general-purpose models. SB 53 applies only to developers of frontier models and catastrophic risks.
RAISE Act (New York) – New York State has a similar but separate law, signed in December 2025. It largely follows the SB 53 model but has its own rules, such as a shorter 72-hour incident reporting deadline.

Related concepts

Broader

AI regulation, AI safety, transparency

Narrower

security framework, incident report, whistleblower protection

On the edge

EU AI-rendelet, SB 1047, flagship model, catastrophic risk, RAISE Act

A brief history

The bill was introduced by California Senator Scott Wiener after his previous, stricter proposal, SB 1047, was vetoed by Governor Gavin Newsom in 2024. Newsom signed SB 53 on September 29, 2025, and it entered into force on January 1, 2026.

Where things stand

In force since January 1, 2026, it requires major developers to publish their safety frameworks. Other US states regard it as a model, and New York has passed a similar law. Meanwhile, there is a debate over how far federal rules can override state AI laws: in December 2025 an executive order was also issued to curb state AI rules.

Status: 2026 November

Also known as

SB 53 · SB-53 · Senate Bill 53 · California SB 53 · Transparency in Frontier Artificial Intelligence Act · TFAIA · California AI transparency law · California AI safety law

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.