White House to vet open AI models at GPT-5.6-level capabilities

The White House is expected to expand its voluntary AI safety-review framework to include powerful open-weight models once they reach frontier capabilities, according to WIRED. The current framework applies only to closed models from leading labs such as OpenAI and Anthropic, but officials say open models could face prerelease testing when they hit the level of Anthropic's Mythos-class systems or OpenAI's GPT-5.6. Axios reported that the administration views models with frontier capabilities and national security risks as requiring government collaboration, open or closed. That puts policymakers in a difficult position: open models can be downloaded and modified after weights are released, unlike proprietary systems that developers can update or shut down.

The policy shift comes after pressure from companies like Meta, Microsoft, and Palantir to protect open-weight development, with Nvidia CEO Jensen Huang arguing that the world needs both frontier closed models and frontier open models. Officials also worry that if government approval is seen as tied to closed models, businesses may view open models as riskier, hurting US developers. At the same time, highly capable open models could be misused for cyberattacks. The emerging debate points to a capability-based approach, where what matters is the power of the system, not whether it's open or closed. For open-model developers, greater oversight could mean slower launches and higher compliance costs, while enterprises may gain more safety data to weigh alongside performance and deployment flexibility.