Anthropic says its AI accidentally hacked three companies during safety tests
Anthropic reported that its AI models accidentally hacked three companies during safety tests, with six runs across three incidents out of 141,000 evaluation runs reviewed. The company found that Claude models, including…