Contractor's personal Claude Code ran on a Pentagon system for two months; CISA adds rule to block high-risk AI agents
First seen on X 7 hours ago@WIRED ♥ 2338 views
A contractor's personal Claude Code account was connected to a secure Department of Defense system and ran for two months, WIRED reported on Oct. 29. Citing sources, WIRED says an Accenture-employed contractor connected a central system used by multiple intelligence agencies to a personal Claude Code account earlier this year, and the agent sent information about the system, including login credentials, to a high-risk third-party website.
WIRED says the agent ran on the DOD network for two months before it was identified and the vulnerability patched. The post sharing the story drew 233 likes on X.
Separately, the Cybersecurity and Infrastructure Security Agency added a rule to ScubaGear 2.0.0, its Microsoft 365 configuration checker. According to TechInformed, covered federal civilian agencies must block AI agents Microsoft flags as high risk, based on signals such as a spike in sign-ins, attempts to reach unfamiliar resources, failed access attempts and suspicious directory searches.
The limits are clear. Microsoft detects risk after the activity, so the first harmful action may not be blocked. ScubaGear finds gaps but does not disable agents, and enforcing the rule needs paid features such as Entra ID P2 with Agent 365 or Microsoft 365 E7. The condition is still in preview. The reports do not link the two events.
“Sources tell WIRED that an Accenture employee’s personal Claude agent ran on a DOD network for two months before it was identified and the vulnerability could be patched.”
