Army Cyber Command is developing an AI platform designed to move through other networks on its own and neutralise what it finds there, according to internal documents obtained by The Intercept.
The program is codenamed ECHOTRIBBLE. A paper prepared for potential contractors describes a system built to navigate infrastructures autonomously and identify and neutralise threats and targets.
Most offensive cyber work today is done by soldiers at keyboards. A vendor question-and-answer document states the effort is primarily focused on attacks.
A worm with a target list
The architecture described is worm-like. Agents jump between devices undetected until they reach the intended target, including across cloud environments and air-gapped systems with no network connection.
Neutralising a target could mean disabling a threat completely or temporarily, or deleting it from the machine. Once the job is finished, the agents are meant to erase themselves.
The Army can set predefined triggers up front, but the documents say the software must operate with minimal human oversight once released.
That is where outside experts see the danger. A self-propagating agent does not reliably distinguish between the network it was sent into and the networks adjacent to it, and collateral damage in cyberspace is difficult to recall once code is moving.
Self-replicating software has caused unintended spread before, from the Morris worm onward. Adding a language model to the payload changes the judgement inside the loop, not the physics of propagation.
