“It was one of the most serious things we saw in her chats,” he said. “Even when we saw things about getting support, sycophancy started.”
OpenAI has previously stated that it has “a deep responsibility to help those who need it most.” The company did not immediately respond to Ars’ request for comment on the new case.
“Our goal is to make our tools as useful as possible to people — and as part of that, we’re continuing to improve the way our models recognize and respond to signs of mental and emotional distress, and connect people with care guided by expert input,” the company wrote in August 2025, less than two months after Carrier’s death.
Earlier this year, OpenAI said that the ChatGPT-4o model specifically would be retired (it has been retired once before, then brought back).
Brown told Ars that he’s not entirely convinced the problem of potentially lethal sycophancy has been solved.
“I think we believe the company has taken steps in the right direction,” he said.
“We don’t have confidence in how security mechanisms are being implemented and how security teams are being implemented and heard,” Brown said. “Of course, we’ve heard that OpenAI has done a lot of things and put out a lot of blog posts and statements related to rolling things back and taking security measures. But ultimately it should have been done sooner. These products are typically brought to market very quickly.”
If someone you know is suicidal or in crisis, please call the Suicide Prevention Lifeline at 1-800-273-TALK (8255), which will put you in touch with a local crisis center.
<a href