Meta contractors posed as teenagers while testing rival chatbots with prompts about suicide, sex and drugs. The testing focused on how competing systems responded to sensitive minor-perspective scenarios.
The report raises difficult questions about safety evaluation, competitive intelligence and the ethics of simulating vulnerable users. Testing high-risk chatbot behavior is necessary, but how companies run those tests matters.
For AI platforms, the story underscores that child safety and crisis-response behavior remain central trust issues for consumer assistants.