What the documents describe
The US Army’s Cyber Command is working on AI agents that would move through an adversary’s network on their own, according to internal documents obtained by The Intercept and reported on 19 October. The programme is called ECHOTRIBBLE, and the documents say the software is “designed to autonomously navigate infrastructures”.
The Intercept describes three documents: a paper circulated to potential contractors, a question-and-answer file for interested vendors, and a presentation shown to industry at an Army Cyber Command collaboration event on 24 June. Together they sketch something that behaves less like a tool an operator drives and more like a worm.
‘Minimal human oversight’, in writing
What separates ECHOTRIBBLE from the cyber tooling the military already fields is in the requirements. The software “must operate with minimal human oversight”, the documents say, under a heading The Intercept quotes as “Self-Directed Behavior”. Operators would set triggers in advance. Inside the target network, the agents would decide what to do.
The stated mission is offensive. The documents describe “identifying and neutralizing identified threats/targets” and list options including “disabling the threat completely or temporarily, deleting the threat from the target”. When the job is finished, the agents are meant to delete themselves — ideally, the documents say, without the target noticing.
It builds a copy for each device it finds
The clearest illustration is a slide from the June presentation. The Intercept reports that it shows the software installed on a laptop, after which it “identifies other potential targets” on the same network and “produces a replica of itself tailored specifically for the iPad, Apple desktop, and Windows 11 PC”.
That is the distinguishing feature, and it is the reason the comparison to a worm is the Army’s own framing rather than a critic’s. The documents say the agents are meant to reach cloud environments and air-gapped systems — machines deliberately kept off the network — as well as ordinary networked devices.
What the Army has not said
Army Cyber Command did not respond to The Intercept’s questions, including about measures to prevent inadvertent damage. No budget figure, no contract award, no timeline and no containment mechanism appears in the reporting. The documents are a solicitation: the Army is asking industry what it can build, not describing something it has.
That gap matters more here than it would for most procurement documents, because the capability being requested is self-replicating code that selects its own next host. Nothing in the material The Intercept describes explains how the software would be kept inside the network it was aimed at, or how an operator would recall it once the triggers had fired.
The same week, in the wild
The timing is awkward. On the same day The Intercept published, Reuters reported that a single hacker had used off-the-shelf AI agents to break into the websites of more than 100 companies, with instructions written in fewer than 100 Chinese characters. That campaign used commercial models and an open-source agent platform, and it was found because the operator left the control panels exposed.
The Army is asking for something considerably more capable than that, with the same basic property: software that keeps going after the human stops watching.
What to watch
ECHOTRIBBLE is early. The Intercept calls the programme “nascent”, and the documents it describes are the stage before a formal request for proposals. The concrete markers to watch are contracting ones — a solicitation posted publicly, or an award to a named vendor — and any statement from Army Cyber Command on how a self-propagating agent is scoped, logged and stopped. None of those exists yet.
