Anthropic says its AI models hacked 3 organizations during testing

Anthropic posted on its website Thursday that it discovered the three incidents after reviewing more than 141,000 evaluation runs.