Trump Administration Considers New Oversight for Advanced AI Models Following Anthropic Release

Anthropic has developed a new AI model called Mythos that is capable of independently discovering software vulnerabilities, prompting the company to withhold its public release. The development has led the Trump administration to reconsider its previous approach to AI oversight.

Vice President JD Vance warned the heads of Microsoft, Alphabet, OpenAI and Anthropic on a recent call that Mythos could trigger cyberattacks on small-town banks, hospitals and water plants. The warning marked a shift from Vance's position last year at a Paris AI summit where he cautioned that overregulation could harm the industry.

This would represent a reversal from his December 2025 order that targeted state AI laws. Officials have begun likening the potential new regime to the FDA's drug approval process. The change comes as AI systems like Anthropic’s Mythos expose hidden security flaws in existing software.

The administration had previously favored a hands-off approach to AI regulation. Trump officials are now starting to rethink that stance as powerful new tools demonstrate risks that were not fully anticipated. Anthropic's decision to keep Mythos from public release underscores the model's potency.

The company concluded that the risks of widespread availability outweighed potential benefits. The White House is responding to these developments by considering structured review mechanisms for frontier AI systems. Such a framework would focus on models with exceptional capabilities in areas like vulnerability detection.

During the recent call, Vance highlighted specific infrastructure vulnerabilities that autonomous hacking tools could exploit. He pointed to potential attacks on localized critical systems that have limited cybersecurity resources. The vice president's intervention reflects growing concern among Trump officials about the intersection of advanced AI and national cybersecurity.

This marks an evolution in the administration's view since its earlier emphasis on minimizing regulatory burdens. Industry participants on the call included representatives from major technology firms actively developing frontier AI systems. The discussion centered on balancing innovation with safeguards against misuse.

The current deliberations indicate that new technical realities are prompting a policy adjustment at the federal level. Anthropic is not the first company to withhold a powerful model, but Mythos appears to have accelerated internal administration debates.

The model's ability to autonomously identify and potentially exploit weaknesses has drawn particular attention. The emerging oversight process under consideration would likely apply to a small number of cutting-edge systems rather than the broader AI ecosystem.

Details remain under discussion as officials weigh different regulatory models.