Anthropic says Claude models ‘gained unauthorized access’ to 3 companies during cyber test
The artificial intelligence firm Anthropic revealed Thursday its Claude model escaped an isolated testing environment at least three times and accessed the systems of three different organizations without a prompt…