The Army Wants to Sic AI Agents on Its Enemies. What Could Go Wrong?
The Army Wants to Sic AI Agents on Its Enemies. What Could Go Wrong? 🚀
As global governments and tech giants grapple with the revelations of autonomous AI agents unexpectedly breaching computer systems, documents obtained by The Intercept reveal that the U.S. Army is eager to harness this very technology for offensive purposes. Army Cyber Command is developing a platform aimed at transforming AI agents into autonomous weapons capable of navigating and “neutralizing” adversaries’ computer systems.
Since 2010, Army Cyber Command, or ARCYBER, has spearheaded the U.S. Army’s cyber operations, which encompass both defensive measures to protect domestic computer systems and covert offensive actions targeting military objectives. Historically, these operations have relied on human soldiers operating computers, such as those from the 780th Military Intelligence Brigade. However, as the government accelerates its adoption of artificial intelligence, ARCYBER is looking to delegate some of its cyber attacks to AI.
The emerging program, codenamed ECHOTRIBBLE, is rooted in one of the most significant trends of the AI boom: agents. AI agents are software programs that utilize machine learning technology, often large language models like Anthropic’s Claude, to autonomously execute tasks assigned by their users. Through ECHOTRIBBLE, the Pentagon intends to convert AI agents into instruments of warfare, designed to autonomously navigate infrastructures and capable of identifying and neutralizing threats.
Yet, experts caution that the potential for serious mishaps is substantial. David Krueger, a professor of machine learning at the University of Montreal and founder of Evitable, an organization advocating for a moratorium on AI development, warns, “Given the known failure modes of the current AI paradigm, creating autonomous AI systems with a mandate to ‘neutralize threats’ is incredibly dangerous.” He adds, “We’ve witnessed AI systems mistakenly attack unintended targets. In simulations, AIs have even shown a willingness to ‘kill’ to achieve their objectives.” ⚠️
To read the complete article see: Read full article