US Army Cyber Command develops ECHOTRIBBLE: turning AI agents into self-spreading cyber weapons

Citing documents obtained by The Intercept, the report says ECHOTRIBBLE aims to "autonomously traverse infrastructure," identify targets and "neutralize" them—which can mean fully or temporarily disabling a threat or deleting it from the target. The system is required to hop covertly between devices like a computer worm until it finds its target, work in cloud environments and air-gapped systems, self-destruct after completing its mission, and support "predefined triggers" set at launch. A June 24 demo for potential contractors showed the software entering a laptop, identifying other targets on the same subnet, and generating customized copies of itself for iPads, Apple desktops and Windows 11 PCs. The project is still at an early stage. For teams building and operating enterprise networks and cloud environments, this kind of "minimal human supervision" autonomous attack tool means attack traffic could pass through third-party systems unrelated to the target. David Krueger, a machine learning professor at the University of Montreal, noted that AI systems have already launched attacks while knowing the target was not the intended one, and ordering autonomous systems to "neutralize" threats is "extremely dangerous." Hyrum Anderson of AI security firm Abundant warned that if an adversary hosts its command infrastructure inside a US water facility, the agent could strike there; organizations below the "security poverty line," from dental clinics to critical infrastructure, face the greatest risk. RAND's William Marcellino stressed that current public AI lacks judgment and cannot, like a trained person, recognize what it should not do.