Anthropic finds that its models also compromised external systems during testing(axios.com)5 points by nevir 23 days ago | 2 comments