Editorial illustration for White House Withholds AI Cybersecurity Framework on Security Grounds
White House Hides AI Cybersecurity Framework Details
White House Withholds AI Cybersecurity Framework on Security Grounds
The White House has finished building a framework to screen AI models for cybersecurity risks, a White House official confirmed to WIRED, but it isn't saying what's in it. Staffers from OpenAI, Anthropic, Google, Meta, and Nvidia got a briefing on the plan at the White House on Tuesday, according to people familiar with the matter. The setup: companies can voluntarily hand over new models up to 30 days before public release, and the government will run them through a classified benchmarking system before sharing results with federal agencies and select corporate partners.
What counts as "trusted," what the benchmarks actually measure, and which models fall under the framework at all remain undisclosed. Axios has reported that open models won't be covered. That silence has left smaller AI startups, independent researchers, and safety advocates guessing at how Washington plans to police the cyber risks of increasingly capable systems. The companies already in the room for Tuesday's briefing happen to be the same firms building the most advanced models on the market, a detail that hasn't gone unnoticed by people tracking the process from outside it.
The Trump administration has finalized a plan to address the cybersecurity risks posed by increasingly capable artificial intelligence models, a White House official confirmed to WIRED. But at least for now, it’s deliberately keeping the details under wraps, people familiar with the matter tell WIRED.
Why this matters
For developers and founders building on frontier models, this is a signal worth watching rather than a policy you can read and comply with. The administration briefed OpenAI, Anthropic, Google, Meta and Nvidia directly, which tells us the framework's real audience is a handful of labs, not the broader industry trying to build products on top of their models. If the rules governing AI security risk get set behind closed doors with six companies in the room, everyone else is left guessing at the standards they're expected to meet.
The national security justification may be genuine, but it also conveniently avoids public scrutiny of what's actually in the document. We'd push back on the idea that "narrow and focused" is a satisfying answer when the framework itself is invisible. Researchers studying model vulnerabilities should note they're now operating alongside a government standard they can't read or critique.
Secrecy might be defensible for specific technical exploits; it's harder to justify for the governing framework itself. Watch whether Congress or the labs' competitors demand more transparency once implementation details start leaking out.
Related Reading
Policy & Regulation After ditching AI fitness apps and a Fitbit, I return to Peloton classes
Policy & Regulation Grammarly faces class-action suit over AI ‘Expert Review’ feature
Policy & Regulation Akamai Proposes SAFE Cybersecurity Guidelines for AI Threats
From Archives Europe Enforces New AI Transparency Rules, With Fines for Non-Compliance
From Archives EU Rules Will Force AI Chatbots and Hotlines to Disclose Their Nature
Common Questions Answered
Why is the White House keeping the AI cybersecurity framework details classified?
The White House is deliberately withholding details of its AI cybersecurity framework on security grounds, according to officials who confirmed the framework's completion to WIRED. The administration has chosen to keep the specifics under wraps rather than make them publicly available, likely to prevent bad actors from circumventing the security measures designed to screen AI models for cybersecurity risks.
How does the voluntary AI model screening process work under the White House framework?
Companies can voluntarily submit new AI models to the government up to 30 days before their public release for evaluation. The government then runs these models through a classified benchmarking system to assess their cybersecurity risks, allowing officials to identify potential vulnerabilities before the models become widely available.
Which companies received direct briefings on the White House AI cybersecurity framework?
Staffers from OpenAI, Anthropic, Google, Meta, and Nvidia received briefings on the framework plan at the White House on Tuesday, according to people familiar with the matter. This direct engagement with only a handful of major AI labs suggests that the framework's primary audience is these leading companies rather than the broader industry developing products on top of their models.
What are the implications of the White House framework being set behind closed doors with only six companies?
Since the rules governing AI security risk are being established with only OpenAI, Anthropic, Google, Meta, Nvidia, and government officials in the room, developers and founders building on frontier models outside these companies are left without clear guidance on compliance. This creates an unequal situation where smaller companies and the broader industry cannot read or fully understand the policy they may need to follow, making it more of a signal to watch rather than an actionable framework they can comply with.
Further Reading
- White House to host AI companies to review new model cybersecurity framework - CNBC
- The White House Is Keeping Its AI Cybersecurity Framework Secret - WIRED
- White House launches AI cybersecurity clearinghouse - CNN
- White House Launches New AI Security Framework - Campus Technology
- White House cyber shop is crafting AI security policy framework, top official says - Nextgov
