The cybersecurity agency’s use of the AI model comes as Anthropic’s relationship with Washington remains anything but smooth.
CISA has been feeding government code through Anthropic’s Mythos model to hunt for security flaws, according to three people with direct knowledge of the effort. The scanning is reportedly turning up a substantial number of vulnerabilities.
What’s happening
The work is being carried out by CISA’s Attack Surface Evaluation team, a unit that regularly runs security assessments and hacking drills across federal systems. Two sources say the audits have already flagged numerous bugs, though the exact scope – how much code was reviewed or how serious the flaws are – remains unclear. Neither Anthropic nor CISA has commented publicly on the initiative.
Why it matters
This is the latest twist in a bumpy year for Anthropic and the federal government. Tensions peaked in February when the Pentagon hit the company with a rare supply-chain risk label after it refused to strip safeguards blocking military and surveillance use of its AI. A judge blocked that designation in March.
Since then, ties have thawed somewhat, helped along by Mythos, a model reportedly built to be unusually skilled at both finding and exploiting security holes. The NSA has apparently been quietly using it since April, and Fable, the public version of Mythos, was briefly pulled offline worldwide last month before access was restored last week.
What to watch next
Traders should watch for any official comment from CISA, Anthropic, or the White House, and for signs of how this affects Anthropic’s IPO timeline.
Stay ahead of every market-moving headline with QuoMarkets.
Source: Reuters
