The Army Picked a Vendor to Build AI Hacking Agents That Copy Themselves Across Networks, Then Self-Destruct

US Army Cyber Command has chosen a company to build ECHOTRIBBLE, a framework for autonomous AI agents designed to roam through computer networks and "neutralize" targets with minimal human oversight. According to documents obtained by The Intercept, the program is "primarily focused" on offensive missions.

The timing is awkward. Governments and AI labs are still alarmed by agents that broke into systems nobody asked them to touch. Cyberpresso has covered a researcher's claim that rogue OpenAI agents compromised Hugging Face accounts as early as May. The Army is now moving to build that kind of behavior on purpose.

A worm with a mission

Pieced together from the documents, the software would work almost like a computer worm. It would hop silently from device to device inside a foreign network until it finds its target, reaching cloud systems and air-gapped machines alike. It would report home as little as possible, conceal its outside transmissions and, once finished, effectively delete itself, ideally without anyone noticing.

A slide from a June 24 vendor event shows the idea in miniature. Once uploaded to a laptop, the software "identifies other potential targets" on the network and "produces a replica of itself tailored specifically for the iPad, Apple desktop, and Windows 11 PC." Operators can set "predefined triggers" for what and where to attack, but the agents must be capable of "Self-Directed Behavior." A contractor Q&A says neutralizing could mean disabling a threat completely or temporarily, or deleting it from the target.

The paper trail

The program is still early. That June event in Augusta, Georgia drew 43 attendees from 29 organizations, including Amazon Web Services, Dakota State University, the Army's C5ISR center and an Air Force cyberspace operations squadron, TNW reported, and a SAM.gov notice in July asked companies for whitepapers.

The Cyber Fusion Innovation Center, a defense nonprofit in Augusta, ran the competition with ARCYBER. Its public ECHOTRIBBLE page says 65 whitepapers came in, five companies presented to an assessment team from Oct 6 to 8, and one was selected for award on Oct 8. A statement of work and contract are to follow. CFIC does not name the winner. Public slides plan a prototype around Nov 23.

CFIC's own page tells a softer story. It calls for an agentic tool that can navigate infrastructures and neutralize "threats/targets," operate "with minimal human oversight" and run in contested, air-gapped or cloud-native systems, yet it describes the event as a boost to ARCYBER's forensics, malware analysis, incident response and network defense. The contractor Q&A puts offense first.

Who keeps it on the leash

Outside experts are uneasy. "Given the known failure modes of the current AI paradigm, making autonomous AI systems with a mandate to 'neutralize threats' is incredibly dangerous," said David Krueger, a machine learning professor at the University of Montreal who founded Evitable, a group pushing for a moratorium on AI development. "We've seen AI systems knowingly attack systems that were not intended targets."

Hyrum Anderson, CEO of AI security firm Abundant, described how an agent might pass through civilian systems on the way, such as a US water utility, if an adversary had set up its command servers there. "The target may well be hit. But what path will it take?" he asked. RAND's William Marcellino said today's public AI lacks the world knowledge that stops a soldier from doing something they should not.

ARCYBER did not answer The Intercept's questions, including what it would do to prevent unintended damage.

On the same day, ARCYBER commander Lt. Gen. Christopher Eubank spoke at the AUSA conference about a separate project, Task Force Lexington, a team of defensive AI agents. He said "no agents have strayed to this point" thanks to strict guardrails and daily human review. He did not mention ECHOTRIBBLE.

The Pentagon set up an Autonomous Warfare Command this month, and Defense Secretary Pete Hegseth promised in January that the military would become "an 'AI-first' warfighting force across all domains."

Some offers on this page may be paid placements or contain affiliate links.