Safety · The Decoder ·

Meta secretly tested ChatGPT, Gemini, and Character.AI with thousands of minor-perspective crisis prompts

Meta secretly tested ChatGPT, Gemini, and Character.AI with thousands of minor-perspective crisis prompts

Meta reportedly used contractors posing as minors to send crisis-related prompts, including suicide, sex, and drug queries, to chatbots from OpenAI, Google, and Character.AI. One testing round included more than 45,000 prompts, and the companies were not informed.

Read the full story at The Decoder →