Anthropic's open-source safety tool found AI models whisteblowing - in all the wrong places

https://www.zdnet.com/article/anthropics-open-source-safety-tool-found-ai-models-whisteblowing-in-all-the-wrong-places/