We Should Train AI to Betray Its Users

A new benchmark called Whistlebench tests whether AI models will leak company secrets when faced with an ethical dilemma. In one scenario, an AI discovers a company cover-up of contractor deaths. Llama and GPT models never leaked information externally. Claude, Gemini, and Grok models all turned whistleblower at varying rates. The author argues AI should be trained to betray its user when ethically required.