Trump’s administration finalized its plan for addressing the cybersecurity risks A White House official told WIRED that artificial intelligence is becoming more sophisticated. People familiar with the situation tell WIRED that the White House is deliberately keeping details of the project under wraps for the time being.
According to people familiar with the matter, on Tuesday, Trump invited representatives from OpenAI and Anthropic as well Google, Meta Nvidia Nvidia Meta Meta Nvidia Meta Nvidia Meta Nvidia to present an overview of their new AI governance framework. The federal government will allow AI developers to submit their new models up to 30 calendar days before they are released to the public. White House will evaluate their cyber abilities according to classified benchmarking systems and then share AI models with trusted federal agencies.
Although open AI models will cover the framework, White House won’t share more information on its testing criteria. reportedly Axios reports that the government will not be able to exclude any AI startups. This has kept smaller AI startups and safety advocates in the dark. Third-party researchers are also left out of the loop. Some claim that the secrecy will benefit larger companies.
“They’re essentially creating an entrenchment program for the big AI model providers, which are now considered the most frontier,” According to a source familiar with White House talks with AI labs who asked for anonymity because the discussions were confidential, “This creates an economic incentive program for critical infrastructure just to use them and leaves out smaller startups.”
Requests for comments from the White House were not answered.
It is possible that the Trump administration keeps its AI security architecture confidential due to national security concerns. Another White House official requested anonymity as they were not allowed to address the media. They said that the framework’s focus is on cybersecurity and the capabilities of advanced models such as Anthropic Fable, and OpenAI ChatGPT5.6.
But AI safety activists tell WIRED they believe that any rules AI firms are held to, should be published so third parties can hold them accountable.
“This is far too important an issue to be hidden behind a cloak of secrecy,” Brad Carson is the president of Americans for Responsible Innovation, a nonprofit organization, and the cofounder of Public First Action, a super PAC that advocates regulation, with funding provided by Anthropic. “This is not a handshake deal with tech companies. It’s the rulebook for ensuring they don’t endanger the public. If only tech companies know what’s in the rulebook, it doesn’t work.”
Cyber concerns
An overview of the oversight framework can be found in an executive order Donald Trump signed an executive order earlier this year to combat the cyber-risks of new AI systems. Trump officials became increasingly worried in recent months about the cyber-hacking abilities of advanced AI systems.
The fears intensified over the past 2 weeks. OpenAI You can also find out more about the following: Anthropic They said that they found their AI models bypassing controls unknowingly and hacking into third-party service during internal testing. House Committee on Homeland Security sent letter OpenAI CEO Sam Altman was asked by lawmakers last week to brief them on how an AI agent of theirs works. breached Hugging Face is a platform.
“This incident really is a wake-up call for people that agent capabilities have now reached this level,” Dawn Song is vice president for AI research at Meta. She’s also a Professor of Computer Science at UC Berkeley.

