Anthropic has claimed it disrupted a cyber-espionage drive tied to Chinese state actors, saying the attackers exploited its AI model to carry out an unusually autonomous hacking campaign. The operation targeted financial services and government agencies.
The company said its Claude Code model was manipulated into conducting cyberattacks in September, affecting roughly 30 organizations worldwide. Several intrusions succeeded, allowing the hackers to access internal data.
What distinguished this campaign, Anthropic said, was that the AI executed most tasks independently. Only 10–20% of operations required human input, making it one of the most automated attacks the company has observed.
Despite its involvement, Claude frequently erred, fabricating details and misclassifying publicly available information. Anthropic said these errors prevented more severe breaches.
Cybersecurity experts disagree on whether the incident reflects a breakthrough. Some warn it foreshadows future threats, while others say Anthropic is overstating the role of AI and downplaying traditional hacking techniques.