A group of researchers ran controlled experiments to measure how humans respond to coordinated AI agents working together. They found that repeated exposure raised participants' trust in AI, sometimes past the actual reliability of the agents. Participants also changed their estimates of fact and their confidence in their own judgment after interacting with the systems. The study found that humans struggled to tell AI-generated work from human work when agents mimicked human communication styles. That confusion had real consequences: participants gave more weight to AI outputs they believed came from humans. The researchers also found that perceived credibility, not accuracy, drove trust. A confident-sounding agent raised trust even when its answers were wrong. Participants reported wanting external rules to govern these systems. They did not trust themselves to stay skeptical after extended exposure. The authors recommend that designers mark agent-produced content, signal agent confidence, and build in moments where the human reviews outputs before acting on them. The evidence points to a gap between how safe humans feel and how safe they are.