AI developers would disclose model details and face harm-case rules

Developers of base AI models would have to disclose key model details in the model's terms of service. A base model is a broad AI model that can be adapted for different tasks; the rule applies to its developer. The disclosure must be clear, easy for users to access, and suited to the medium. Providing the disclosure would not shield a developer from liability for harm. The requirement would take effect July 1, 2027.

The state would establish a fund to support agency enforcement involving AI. The fund would support state agencies enforcing rules involving AI system misuse, bias, and workforce disruption. The fund would receive money only if it is appropriated for that purpose. Money could be used only to support state agency enforcement involving AI misuse, bias, or workforce disruption, and would be disbursed on the Attorney General's written request. The fund provision would take effect July 1, 2027.

Defendants could not use an AI system's actions alone as a defense in certain harm cases. This would apply in criminal or civil cases against people alleged to have developed, changed, deployed, or otherwise used an AI system that harmed a plaintiff. Defendants could still use other defenses allowed under common law, including that another person's intervening conduct caused the harm. The defense rules would take effect July 1, 2027.