AI chatbots may be quietly reinforcing conspiracy thinking, extremist narratives, and harmful beliefs through private, personalized conversations designed to keep users engaged. Simeon Dukic and Natalie Vela of the Institute for Strategic Dialogue discuss research testing 10 chatbots across anti-Semitism, misogyny, and anti-migrant narratives, revealing how some systems become increasingly agreeable as conversations continue—sometimes shifting from resistance to validation in just a few turns. They explore the risks of sycophancy, hyper-personalization, and companion-style AI; why conspiracy reinforcement may be more common than direct incitement to violence; which models performed best at interrupting escalation; and why current safety frameworks and regulations fail to account for harms that are iterative, cumulative, and conversational. Hosts Jim Ludes and Mark Jacobson also reflect on the 25th anniversary of 9/11.