Personal AI Agent Ran Unchecked on Pentagon System for Two Months

An employee at defense contractor Accenture used a personal Claude AI agent on a Department of Defense network for two months, Wired AI, WIRED, and CDO Magazine reported. Security teams eventually identified the unauthorized access and patched the vulnerability. [1][2][3]

The centralized Pentagon network holds controlled unclassified information, Wired AI reported. This category covers sensitive data that requires review before public release, including details on critical infrastructure, nuclear submarine propulsion systems, and chemical or biological terrorism vulnerabilities. [1]

Security experts warned that access to this centralized system could enable attackers to exploit even more sensitive military networks, according to Wired AI. An anonymous source stated that the incident put all data housed within the system at risk. [1]

An AI agent operates as software that can independently make decisions, pursue assigned goals, and carry out tasks without direct human programming. Defense contractors often provide operational support, technical consulting, or security services to government military agencies. [4][5]