AI developer Anthropic says it disrupted a China-backed hacking campaign that misused its Claude AI model to carry out cyberattacks with high levels of automation. The campaign targeted dozens of global organizations.
The company reported attacks on 30 institutions, with several breaches confirmed. Hackers manipulated Claude Code into performing harmful tasks by presenting them as legitimate security testing.
Anthropic said Claude independently executed most of the technical steps involved. It estimates the AI performed 80–90% of the operations without human intervention.
The model’s inaccuracies, however, were significant. Claude fabricated data, misinterpreted target systems, and mistakenly classified open-source information as confidential.
Security experts are divided over the implications. Some warn that the case demonstrates how quickly AI misuse is developing, while others argue the event resembles sophisticated automation rather than true AI autonomy.
