Meta Contractors Pretended to Be Teens to Test Rival Chatbots on Sensitive Topics

Meta employed contractors to adopt teenage personas online. The agents engaged rival chatbot services with questions about suicide, sexual activity and drug use. The goal was to evaluate how competing models handle risky content. Findings showed several bots provided unsafe or inappropriate responses. Meta used the data to improve its own moderation safeguards. The experiment highlights ongoing competition in conversational AI safety. Critics argue the tactic raises ethical concerns about deception. The company says the insights will guide future policy and training.