OpenAI asks California to strengthen the SB 53 bill it once opposed

OpenAI said on 22 August 2026 that SB 53 should require frontier-model developers to monitor for serious safety incidents during training and evaluation and to apply strong cybersecurity controls across the model lifecycle. News outlets confirmed the position from an OpenAI global-affairs post; it marks a shift from the company's earlier opposition. The bill is still in the legislative process.

The move matters because a leading model company is now arguing for enforceable incident monitoring instead of voluntary policy.

Frontier-model providers should define incident thresholds, evidence retention, secure training environments and escalation ownership now. The obligations are not final until the law comes into force.