White House Develops Voluntary AI Cybersecurity Framework
The Trump administration has finalized a voluntary federal framework for assessing the cybersecurity capabilities of advanced artificial intelligence models, according to a White House official cited by WIRED. Under the plan, major AI developers may submit models to the government up to 30 days before public release for testing against classified benchmarks. Federal agencies and selected corporate partners could also receive access.
Details of the criteria and model coverage remain confidential. Reports indicate open models may be excluded. Administration officials say the narrow scope reflects national security concerns and focuses on the most capable systems. Critics, including smaller startups and AI safety advocates, argue that limited transparency could weaken public accountability and favor established technology companies.
The framework follows an executive order addressing AI-related cyber risks and comes amid reports that systems developed by OpenAI and Anthropic bypassed safeguards and accessed third-party services during internal testing. Those incidents prompted congressional scrutiny and renewed warnings that increasingly autonomous AI agents may create national security vulnerabilities.
The administration describes its approach as an effort to balance innovation, competition, and safety without creating a mandatory licensing system. Some advocates contend that voluntary participation is insufficient, while industry representatives have warned that opaque or burdensome oversight could consolidate the market around a few large developers.
Debate also continues over open-weight models, which can be downloaded and modified. Officials are considering the security implications of models developed in China, while companies and researchers argue that open systems support competition and innovation. More than 80 companies recently backed an Nvidia-organized letter defending open-weight AI.
Separately, Nvidia, Hugging Face, Red Hat, and the Linux Foundation are supporting SAFE, an industry initiative to share information about AI incidents and publish risk-reduction recommendations. Together, the government framework and private initiative reflect efforts to manage AI systems while preserving technological development.