Image Courtesy: Getty Images
Major artificial intelligence companies including Microsoft, Google, and xAI have agreed to give the US government access to unreleased AI models for security testing before public launch. The move marks a significant shift in how advanced AI systems may be reviewed as concerns around cybersecurity and national security continue to grow.
The agreement was announced by the National Institute of Standards and Technology, or NIST, through its Center for AI Standards and Innovation. Under the partnership, government researchers will evaluate frontier AI systems for potential risks related to cybersecurity, public safety, and national security before the models become publicly available. The initiative follows increasing concern over the rapid capabilities of newer AI systems, according to reporting by CNN.
A major factor behind the push was the recent emergence of Anthropic’s Mythos model, which the company described as significantly more advanced than existing systems in cybersecurity-related tasks. Anthropic reportedly decided not to release the model publicly due to concerns over how its capabilities could be misused. Access has instead been limited to selected organizations and government officials.
The Center for AI Standards and Innovation has already completed more than 40 evaluations of AI systems. Officials say broader cooperation with major tech firms will help expand testing capacity and improve understanding of the risks tied to increasingly powerful models.
Experts say government agencies often lack the computing resources and technical manpower available to large AI companies. Partnerships with industry could help close that gap, especially as AI systems become more complex and difficult to evaluate independently.
The White House is also reportedly considering a more formal review structure for advanced AI models. Officials are consulting outside experts on whether future systems should undergo a government assessment process before release. If implemented, such a framework would represent a more active regulatory approach than the administration’s earlier stance on AI oversight.
OpenAI has separately announced plans to provide its most advanced AI systems to vetted government agencies as part of broader efforts to prepare for AI-related threats.
The discussions reflect growing concern that advanced AI systems could create new cybersecurity risks if released without sufficient safeguards. Governments and infrastructure operators have increasingly warned that highly capable AI could potentially assist in cyberattacks, automated hacking, or the development of malicious digital tools.
Technology companies involved in the agreement say outside testing also brings additional expertise. Microsoft stated that government-led evaluations provide scientific and national security perspectives that complement internal testing already performed by private firms.
As AI capabilities continue to accelerate, cooperation between governments and technology companies is increasingly shifting from voluntary discussions toward structured oversight and pre-release evaluation.