The White House is expected to issue an executive order proposing a voluntary government review system for advanced artificial intelligence models before they are publicly released, according to CNN.
The proposal would encourage AI companies to provide US authorities with access to their frontier AI systems ahead of launch so officials can assess potential national security and cybersecurity risks.
The move marks a significant shift in the US government’s engagement with artificial intelligence development, particularly as advanced AI systems become more powerful and capable of influencing sectors ranging from defence to cybersecurity.
Under the draft proposal, AI firms would share unreleased frontier models with government agencies before launch. Officials are still debating how long companies would need to provide access.
One version of the proposal reportedly suggests a 90-day review window. However, several industry players are said to be pushing for a shorter timeline of around 14 days.
The review process would reportedly remain voluntary rather than mandatory, though the framework could create pressure for leading AI firms to participate.
Major AI firms, including OpenAI and Anthropic, are reportedly engaged in discussions with the White House regarding the framework.
The discussions come amid increasing concern among policymakers and cybersecurity experts that highly advanced AI systems could be misused to accelerate cyberattacks or expose vulnerabilities in critical infrastructure.
According to CNN, the draft executive order is divided into two primary sections: cybersecurity and “covered frontier models.”
The cybersecurity section reportedly proposes the creation of a voluntary “clearinghouse” involving the Treasury Department, federal agencies and private AI firms. The goal would be to identify and address vulnerabilities in unreleased AI models before they are deployed publicly.
Experts have repeatedly warned that frontier AI systems could help malicious actors automate hacking operations, discover software vulnerabilities faster and scale cyberattacks more efficiently.
The draft proposal reportedly defines a category called “covered frontier models,” referring to advanced AI systems considered powerful enough to require additional scrutiny before release.
These models would fall under the proposed voluntary review mechanism.
While exact technical thresholds remain unclear, the term generally refers to cutting-edge AI systems capable of performing highly complex tasks across multiple domains.
The Trump administration had previously maintained a relatively hands-off approach toward regulating artificial intelligence.
However, recent developments within the AI industry appear to have prompted deeper engagement between government officials and leading AI companies on safety concerns.
CNN reported that one factor influencing discussions was Anthropic’s unreleased “Mythos” model and the company’s controlled access programme, which reportedly drew attention to the potential risks associated with powerful AI systems.
The proposal also reportedly includes plans to expand hiring under the US Tech Force, a government engineering initiative focused on modernising federal IT systems.
The expansion could strengthen the government’s technical capacity to evaluate advanced AI systems and improve cybersecurity preparedness across agencies.
The White House has not officially commented on the reported executive order.
Separately, the Commerce Department’s National Institute of Standards and Technology had earlier indicated that major technology companies would share unreleased AI models for national security evaluations. However, CNN noted that the announcement was later removed from the department’s website.
