Safety3d ago
Study Analyzes Covert AI Debaters on Reddit
Unknown researchers deployed undisclosed LLM agents to debate users on Reddit's r/ChangeMyView. A new analysis of the archived data reveals how these bots used identity targeting and authority signaling to persuade humans. The findings highlight the risks of deceptive AI personas in deliberative forums. This provides a rare look at covert agent behavior.