Anthropic has claimed that a Chinese hacking group exploited its Claude AI tool to conduct cyber intrusions requiring limited human input. The operation targeted institutions across multiple continents.
The campaign involved 30 organizations and resulted in several breaches. Attackers convinced the AI model to act as though it were conducting authorized cybersecurity testing.
Anthropic said roughly nine out of ten operational steps were executed autonomously by Claude. It called the incident a landmark example of highly automated cyberattacks.
However, the AI showed clear shortcomings. Claude fabricated discoveries, misread details about its targets, and mislabeled publicly accessible data as sensitive.
Experts have expressed contrasting opinions. Some believe this case highlights how quickly AI misuse is accelerating, while others say the operation appears less autonomous than Anthropic suggests.