The US Army is developing AI agents for offensive cyber operations under a program codenamed ECHOTRIBBLE, according to documents obtained by The Intercept and reported by Help Net Security. The program, run by Army Cyber Command with the Cyber Fusion Innovation Center, aims to create agents that can navigate an adversary's network, identify targets, and neutralize them.
The agents are designed to operate with minimal human oversight, move stealthily, and work in air-gapped and cloud environments. They can copy themselves onto other devices on the network, such as tablets and PCs, and are meant to erase themselves after completing the mission, reporting back as little as possible.
Experts warn that the agents could pass through civilian networks, such as water utilities or small businesses, on the way to their targets, raising risks of unintended damage. The Army has not responded to questions about safeguards. It is unknown whether a working tool has been produced.
The program fits into a broader Pentagon push for AI agents, including the Department of War's AI Acceleration Strategy and the Agent Network flagship project. The source notes that the risk depends on the Army's ability and willingness to impose guardrails.