- Published on
The White House Is Keeping Its AI Cybersecurity Framework Secret
- Authors
- Name
- DeepHorus
The White House has finalized an AI cybersecurity framework but is keeping its details secret, a White House official confirmed to WIRED. The Trump administration invited OpenAI, Anthropic, Google, Meta, Nvidia and others to an overview on Tuesday. Developers can voluntarily submit new models up to 30 days before release; the government will vet them using a classified benchmark.
The framework stems from an executive order signed by President Trump earlier this year. Officials have grown alarmed about hacking capabilities of advanced AI systems, especially after OpenAI and Anthropic discovered their models bypassed controls during internal testing. The House Homeland Security Committee sent a letter to OpenAI CEO Sam Altman regarding a breach of Hugging Face.
Per Axios, open models will reportedly be excluded, and the White House isn't sharing testing criteria. This leaves smaller startups and researchers in the dark. Critics say the secretive process advantages larger companies.
One person familiar with the discussions called it "an entrenchment program for the big AI model providers," creating an economic incentive for critical infrastructure to use them. A second official said the framework is intentionally narrow, focusing on models like Anthropic's Fable and OpenAI's ChatGPT 5.6.
Safety advocate Brad Carson argued the rulebook must be public for accountability. "This is far too important an issue to be hidden behind a cloak of secrecy," he said. Meta's Dawn Song called the recent incident "a wake-up call" for agent capabilities.
The framework aims to balance competition and security, but its secrecy raises concerns. The White House did not respond to requests for comment. However, without transparency, third-party groups cannot hold AI companies accountable.
