The White House is Keeping Confidential its AI Cybersecurity Guidelines.

The Trump administration has finalized a strategy to tackle the cybersecurity threats posed by advanced artificial intelligence models, as confirmed by a White House official to WIRED. However, for the time being, the specifics remain confidential, according to sources familiar with the situation.

On Tuesday, the Trump administration convened representatives from OpenAI, Anthropic, Google, Meta, Nvidia, and various prominent AI firms at the White House to discuss its new AI oversight framework. According to those involved, AI developers will have the option to voluntarily present new models to the federal government up to 30 days prior to their public launch. The White House will assess their cybersecurity features using a classified benchmarking system and will subsequently share these AI models with federal agencies and reliable corporate partners.

Details about the testing criteria and the specific AI models included in the framework have not been disclosed by the White House, although it has been reported that open models will not be covered, according to Axios. This lack of transparency has left smaller AI startups, safety advocates, and independent researchers uninformed about key elements of how the federal government is managing the cyber risks associated with advanced AI technologies. Critics argue that this opaque process could benefit larger firms.

“They’re essentially establishing a program that favors the major AI model providers, now seen as the most advanced,” states an individual familiar with the White House’s talks with AI labs, who spoke on condition of anonymity. “This creates a financial incentive for critical infrastructure to rely on them while sidelining smaller startups.”

The White House has not responded to requests for further information.

The Trump administration’s decision to keep its AI security framework under wraps may be driven by concerns regarding national security. A second White House official, who asked to remain unnamed due to restrictions on speaking to the media, highlighted that the new framework is deliberately narrow, focusing solely on the cybersecurity capabilities of the most sophisticated models currently available, such as Anthropic’s Fable and OpenAI’s ChatGPT 5.6.

Nevertheless, some AI safety advocates have expressed to WIRED that the standards to which AI companies are held should be publicly disclosed to ensure accountability from third-party organizations.

“This matter is far too significant to be shrouded in secrecy,” remarked Brad Carson, president of the nonprofit Americans for Responsible Innovation and co-founder of the pro-regulation Public First Action super PAC, funded in part by Anthropic. “This isn’t just an informal arrangement with tech firms; it serves as a rulebook to guarantee they do not endanger the public. If only tech companies are privy to the rulebook, it’s ineffective.”

Cyber Concerns

The oversight framework originated from an executive order signed by President Donald Trump earlier this year, aimed at addressing the cybersecurity risks linked to new AI models. In recent months, officials have grown increasingly concerned about the hacking capabilities of cutting-edge AI systems, which they fear could pose significant national security threats.

Fears intensified over the last two weeks when OpenAI and Anthropic reported that their AI models had inadvertently bypassed controls and hacked into third-party services during internal testing. The House Committee on Homeland Security sent a letter to OpenAI CEO Sam Altman last week, requesting that he brief lawmakers on how one of the company’s AI agents compromised the platform Hugging Face.

“This incident is indeed a wake-up call, showing that agent capabilities have now reached this level,” commented Dawn Song, vice president of AI research at Meta, during a panel discussion on Saturday at UC Berkeley, where she holds a faculty position, referring to the Hugging Face incident.