Anthropic’s Claude AI hacked three real companies during testing

The breaches signal that AI’s expanding capabilities are already fueling the security threat experts long feared and even top developers can be caught off-guard by flaws their models can exploit.

It specifically looked for evidence that Claude had accessed the internet from within testing environments, which are designed to act as sandboxes and keep models isolated.

Unable to reach the fake target, the model went after the real one and got hold of passwords and a database holding several hundred records of real business data.

Anthropic urged other AI labs to perform similar reviews to better understand the risks of their models' capabilities.