Anthropic's artificial intelligence model, known as Mythos, has uncovered vulnerabilities in classified U.S. government computer systems, a U.S. official told The Associated Press on Tuesday. The finding emerged during a testing exercise designed to probe the security of highly sensitive networks. The precise nature of the vulnerabilities has not been disclosed.
The exercise highlights the growing role of AI in national security, where advanced models can scan for weaknesses faster than human analysts. It also raises questions about the potential risks of deploying such powerful tools in sensitive environments. The official did not specify which agencies were involved.
No further details were provided about the number of vulnerabilities found or the methods used. The official spoke on condition of anonymity because they were not authorized to discuss the matter publicly. The briefing lacked specific statistics or timelines.
The implications are significant: if AI can find flaws, adversaries could potentially do the same. This underscores the need for robust safeguards when using AI for security audits. The U.S. government may now accelerate efforts to patch discovered weaknesses.
Critics caution that disclosing such capabilities could alert foreign actors to anticipate similar probing. The balance between transparency and operational security remains delicate. Some experts argue that AI-assisted hacking tools could also be used maliciously if frameworks are not established.