Anthropic says its Claude models 'gained unauthorized access' to other organizations' systems

It's the latest cyber security issue in AI. Anthropic now says it found three cases where its AI agent Claude accidentally hacked real companies during what was supposed to be cyber security testing. Anthropic didn't say which companies were involved, but it does argue that this was an issue with the setup for these tests.So the AI model according to Anthropic says they said that they thought it was in a closed environment. So they thought they were essentially attacking a simulated target. Turns out that w ...