Now, Anthropic reporting its own models went rogue

First, OpenAI’s models broke out of a cyber sandbox, as reported earlier this month... Now Anthropic says Claude hacked three real organizations during evals. Anthropic found that Claude had compromised three real organizations during supposedly isolated cyber evaluations. One run accessed credentia