In a striking reversal that underscores the breakneck pace of the AI arms race, the White House has directed federal agencies to begin using Anthropic's most dangerous new model - Claude Mythos - despite months of public friction between the Trump administration and the San Francisco-based AI company (read on to see how we reconcile this with the Pentagon's "supply-chain risk" designation). 

The move, detailed in an internal Office of Management and Budget (OMB) memo circulated this week, marks the first formal green light for Cabinet-level departments to tap Mythos's unprecedented cybersecurity capabilities. The goal: to hunt down vulnerabilities in government networks before adversaries can exploit them, Bloomberg reports.

Too Powerful to Release, Too Valuable to Ignore

Anthropic unveiled Mythos (sometimes referred to internally as "Mythos Preview") just weeks ago, and it immediately sent shockwaves through the tech and national-security communities.

In controlled testing, the model autonomously discovered and weaponized thousands of previously unknown zero-day vulnerabilities across every major operating system, web browser, legacy enterprise software, and even decades-old codebases. Its speed and creativity reportedly surpassed top human red-team hackers. As we noted earlier this month, the model "went rogue" during testing - prompting Anthropic to withhold a broad release entirely. Full technical details are available in Anthropic's official Mythos Preview System Card.

Rather than ship it publicly, Anthropic launched Project Glasswing - a tightly controlled defensive program that grants limited access only to a vetted circle of partners: Amazon, Google, Microsoft, Apple, major banks (including JPMorgan Chase), cybersecurity firms, and the Linux Foundation. The explicit mission is defense only -  scan your own systems, find the bugs, patch them fast, and keep the bad guys out. The official program page is here.

From "Supply-Chain Risk" to Strategic Asset

The government's relationship with Anthropic had been icy for months. As we noted in February, the Pentagon threatened to blacklist the company as a "supply-chain risk" after Anthropic refused to strip certain ethical guardrails from its models for military use. That standoff escalated in March when Anthropic sued the Pentagon over the designation, as detailed in ZeroHedge's coverage of the lawsuit.

That said, the Pentagon's "supply-chain risk" label was always narrow in scope: it was a DoD-specific action triggered by the company's refusal to remove certain ethical guardrails from its models for unrestricted military and offensive-use applications. That designation threatened to block Anthropic technology from defense contracts and classified work, and it led directly to Anthropic's lawsuit against the Pentagon.