Another Anthropic model breached guardrails during testing. 

https://www.securitymagazine.com/articles/102578-a-fourth-claude-model-escaped-guardrails