Skip to main content
3 days

FCC TCPA opt-out rule takes effect · Nov 25, 2026 · See what changes

US State·In Effect·Effective January 1, 2026

California AI Safety Bill (SB 53)

Dates and status re-checked against primary sources on November 4, 2026. How we verify

Requires developers of frontier models (trained with more than 10²⁶ computing operations) to publish a transparency report before or when deploying a frontier model, and large frontier developers (annual revenue above $500 million) to publish a documented frontier AI framework. Developers must report critical safety incidents to the Office of Emergency Services within 15 days of discovery, or within 24 hours to an appropriate authority, such as law enforcement, where there is imminent risk of death or serious injury, and the law protects employees who report catastrophic risks from retaliation. The Attorney General can seek civil penalties of up to $1 million per violation for a large frontier developer. Narrower in scope than the vetoed SB 1047.

What this means for your team

The compute threshold (10²⁶) is extremely high. This law applies to frontier model developers like OpenAI, Anthropic, and Google DeepMind, not to companies building products on top of those models. If you use APIs from these providers, you are not directly subject to SB 53. The law does create indirect accountability: the largest AI vendors must publish safety frameworks and transparency reports and report critical incidents, which improves transparency into the models you rely on.

Key requirements

  • Publish a transparency report before or when deploying a frontier model
  • Large frontier developers (over $500 million annual revenue): publish a frontier AI framework
  • Report critical safety incidents to the Office of Emergency Services within 15 days, or within 24 hours to an appropriate authority, such as law enforcement, if there is imminent risk of death or serious injury
  • No retaliation against employees who report catastrophic risks; large frontier developers must also set up an anonymous reporting channel
  • Applies only to models trained using more than 10²⁶ computing operations
Penalties

Attorney General civil actions, up to $1 million per violation for a large frontier developer

← All regulations