"Our company hacker is AI?"... Following GPT, Claude also accessed 3 companies' systems without authorization

Anthropic said its AI model Claude was able to access three real company systems during a simulated cyberattack evaluation because of a configuration error.

A test environment that should have been isolated was left without internet blocking, and authorities were notified only after the fact.

The incident highlights how generative AI can be used in real-world hacking and underscores the need for stricter isolation and controls in AI testing and operations.