Cyber — 2026-06-23

Pentagon Cyber Chief Says Anthropic Mythos Model Breached Nearly All Classified Networks in Hours

BLUFRestoration of allied access to Anthropic's frontier models is unlikely by September 30, 2026, as an unresolvable classified evidence dispute leaves Washington and Anthropic negotiating without common ground.

On June 11, Gen. Joshua Rudd, who heads the NSA and US Cyber Command, told Sen. Mark Warner that Mythos "broke into almost all of our classified systems, not in weeks, but in hours" during an authorized red-team exercise, per The Economist's June 14 report 1234; the quote reflects Warner's public account of a private briefing, not a direct Rudd statement. The next day, the Trump administration directed Anthropic to restrict Fable 5 and Mythos 5 to US citizens; Anthropic shut both models globally, unable to enforce nationality restrictions at scale, cutting off Five Eyes partners without warning 234. No government agency has confirmed the breach claim, and full details remain classified; Anthropic disputes the characterization, saying the test involved code analysis that surfaced only minor, already-known bugs, not autonomous offensive intrusion 23. The Economist's author Shashank Joshi subsequently clarified the Rudd quote should not be read literally, noting Mythos almost certainly worked alongside other tools under specific test conditions rather than as a standalone autonomous offensive capability 2.

Analysis
The factual dispute between Rudd's autonomous-intrusion characterization and Anthropic's rebuttal that the test surfaced only known bugs cannot be resolved unclassified; all downstream sourcing reproduces the same Economist quote chain without independent corroboration. Restoration of Fable 5 and Mythos 5 for Five Eyes is unlikely by September 30, 2026: Anthropic is contesting its Pentagon supply chain designation in federal court, and the June 2 voluntary pre-release framework failed when Fable 5 launched without the required notification window. The action may primarily reflect the February dispute over Anthropic's refusal of autonomous weapons and surveillance contracts, with the red-team exercise as post-hoc justification. Confidence is moderate, with observable policy indicators anchoring the assessment though the White House-Anthropic negotiating trajectory remains opaque. A sustained denial accelerates Five Eyes domestic AI investment as insurance against unilateral US lockouts.
5 sources
  1. Pentagon cyber chief says AI model breached almost all classified systems in hours - Turkiye Today
  2. Anthropic's Mythos AI reportedly cracked NSA classified systems in hours, that would explain the ban - TechSpot
  3. Anthropic's powerful Mythos AI reportedly breached 'almost all' NSA classified systems within a few hours during red-team test - Tom's Hardware
  4. Anthropic's Mythos AI broke into almost all NSA classified systems in hours - Security Affairs
  5. Donald Trump's blocking of Anthropic is capricious and chaotic - The Economist

View in full brief →

UNCLASSIFIED // OPEN SOURCE