US completes voluntary AI model review framework

The US government said Monday it has finished a voluntary framework for evaluating advanced AI models. The framework aims to give AI developers a structured process to work with the government on whether their models need oversight. The White House didn't share details, but said it's talking with companies like Anthropic, OpenAI, and Google about next steps.

The framework is expected to set rules for confidentiality, cybersecurity, risk management, and IP protection when the government gets models up to 30 days before release. It will also define which trusted partners can access models early. The benchmark for testing advanced cyber attack abilities and the model threshold are classified, but the framework itself isn't. A White House official said not everything non-classified will be public. Sources say a staff-level meeting with companies is planned for Tuesday.