An Accenture contractor’s personal Anthropic Claude AI agent operated on a secure U.S. Department of Defense (DoD) network for two months before it was discovered.
The agent reportedly ran on a system holding sensitive but unclassified data, including information about critical infrastructure. Sources said it may have transmitted intelligence information, system architecture diagrams and access credentials to third parties.
The incident highlights the security challenges posed by autonomous AI agents in secure government and corporate environments. The vulnerability was identified and patched.
Security experts have raised concerns that third-party software used by AI agents could be compromised, potentially creating a national security threat if it is not properly managed.