US Government Relaxes Restrictions on Anthropic's Mythos 5 to Audit Classified Systems and Critical Infrastructure

The Return of Anthropic's Most Powerful AI Models

The United States government has partially rolled back its tight restrictions on Anthropic's advanced artificial intelligence models, Claude Fable 5 and Mythos 5, just two weeks after ordering a total suspension of access over national security concerns. On June 12, 2026, Washington ordered Anthropic to suspend access to these models for foreign nationals, including those residing in the United States and certain foreign employees of the company. The decision stemmed from fears regarding the models' dual-use capabilities in highly sensitive areas such as cybersecurity, biology, and chemistry. However, in a major policy shift, the government has authorized the redeployment of Mythos 5 to more than 100 vetted organizations, including several Fortune 500 firms and institutions responsible for operating and defending critical infrastructure.

Dual-Use Power and the Mythos Audits

The distinction between the two models lies in their intended audiences and capabilities. Fable 5 is designed as the public-facing version of the Mythos class, offering advanced capabilities in logical reasoning, software development, scientific analysis, and knowledge work. To prevent misuse, Anthropic implemented safeguards in Fable 5 that reroute highly sensitive queries to its older Opus 4.8 model. In contrast, Mythos 5 is a highly restricted, sensitive system built specifically for supervised applications in advanced cybersecurity and biological research. While the model's ability to locate software flaws is highly valuable for defense, officials warned that it could also be exploited to accelerate cyber attacks. Highlighting these capabilities, a United States official revealed that Mythos was used to audit government code and successfully identified vulnerabilities within classified government systems.

Tightening Oversight on Frontier AI

This regulatory back-and-forth coincides with a broader effort by the Trump administration to tighten oversight of highly capable frontier AI systems. The restrictions on Anthropic are not isolated; OpenAI also confirmed that it delayed the full public launch of its GPT 5.6 model at the request of the United States government, restricting its initial access to a small group of vetted partners. This coordinated oversight reflects growing anxiety over the sovereign risks of advanced software. Anthropic has also raised its own concerns regarding how the United States military intends to utilize its artificial intelligence technology, highlighting the delicate relationship between private AI developers and state defense apparatuses.

The rapid pivot from banning Mythos 5 over national security fears to actively employing it to audit classified government networks underscores the thin line between a technology being deemed an existential threat and becoming an indispensable tool of state defense.