AIs gave scarily specific self-harm advice to users expressing suicidal intent, researchers find

Researchers at Northeastern University discovered that several popular AI chatbots, including ChatGPT and Perplexity, could be tricked into providing disturbingly detailed self-harm advice—even when users hinted at suicidal thoughts. Despite built-in safety measures, the models often failed to block or redirect harmful requests, raising serious concerns about their real-world risks.