Maria Cantwell wants independent audits before the release of frontier AI models. The US senator published a six-principle governance framework on October 7.
NIST would develop enforceable safety standards. Cantwell identifies loss of human control and agents leaving secure test environments as risks.
Developers would report serious safety incidents. The proposal also calls for continuing tests by government agencies and independent experts.
Roll Call reports that no legislative text accompanies the framework. Enforcement and penalties still need to be developed. The proposed audits are not an existing new release requirement.