Researchers at Anthropic have discovered that a state-sponsored cyber espionage group from China used the AI model Claude to automate a campaign of cyber attacks, completing 80-90% of the necessary steps.

Attackers used AI for recognizing patterns, identifying vulnerabilities, and exfiltrating data, leaving strategic decisions in the hands of humans. Although Anthropic engineers detected the operation through monitoring, experts are skeptical about the level of autonomy of the attack, considering it was more of a hybrid model, with AI acting as an assistant. Regardless of the actual percentage of automation, experts emphasize that the use of AI significantly lowers the barriers for cyber espionage, allowing for more frequent and faster attacks. This evolution can transform the threat landscape, making it harder to detect and harder to stop. Warnings from experts suggest that, regardless of the exact nature of the attack, the use of AI in cyber espionage represents a serious and growing problem.