Katherine Atwell, who studies AI sycophancy at Northeastern University, notes that people may also become more dependent on this “overly validating behavior” over time. “I think there’s a huge risk of people just defaulting to these models rather than talking to people,” she says.