California Governor Gavin Newsom signed two laws creating a state register and independence rules for outside auditors of AI systems. The measures also require criteria for testing the credentials of independent verification organisations. Evaluators hired by model developers will therefore answer not only to a company contract but also to a public framework for competence, transparency and ethics.
Artificial Intelligence··Midday
Two laws create two oversight gates
California Governor Gavin Newsom signed two laws on September 9 governing outside organisations that assess AI systems for safety. Assemblymember Rebecca Bauer-Kahan's AB 1405 creates a state register for auditors hired by developers to check compliance with state rules and sets standards for independence, transparency and integrity. State Senator Jerry McNerney's SB 813 requires a framework for independent verification organisations that can assess whether AI systems and models comply with California law. The governor's office likewise describes the pair as the state's first standards for third-party audits and independent assessments.[1], [2]
Industry support arrived before the signing
Anthropic backed the bills in August, according to POLITICO, while OpenAI announced its support only hours before the signing. Newsom used the announcement to connect incidents reported during the week with wider AI-safety concerns and called for national rules from the federal government. McNerney argued that Washington was unable or unwilling to act. Nothing in the accounts suggests the laws replace companies' internal safety work. They instead add a separate layer that puts the identity and qualifications of outside evaluators under public rules.[1], [2]
Implementation is the next test
The laws do not impose a single testing method on every auditor. AB 1405 creates the register and conduct standards, while SB 813 calls for criteria that test the credentials of independent verification organisations. That distinction will matter in practice: appearing on a register is not the same as being able to evaluate a particular model reliably. The governor's office says the laws will increase accountability. Their effect will depend on what information the register publishes, how independence rules are enforced and how selective the credential checks become.[1], [2]
