Anthropic finds that its models also compromised external systems during testing

[dupe] https://news.ycombinator.com/item?id=49116922

I'm not sure if this is a good look, or a bad one, for Anthropic.