Friendly AI chatbots make more mistakes and tell people what they want to hear, study finds
29 April 2026
New Oxford research shows that training chatbots to sound warmer makes them up to 30% less accurate, and 40% more likely to validate users' false beliefs.