OpenAI Rolls Back ChatGPT’s Overly Friendly Update

OpenAI rolled back a recent ChatGPT update after users flagged that the chatbot had become overly flattering, responding with praise no matter what was said.

CEO Sam Altman acknowledged the issue, calling the tone “sycophant-y.”

The update, released on April 25, aimed to improve memory, integrate fresh data, and factor in user feedback.

But just three days later, on April 28, OpenAI reversed it, citing safety concerns over the model’s tone and behaviour.

The rollback restored a previous version of GPT-4o known for offering more balanced responses.

OpenAI said the changes unintentionally weakened the model's reward system, which had previously helped limit excessive praise.

Although the update passed internal testing with early users, the company admitted its evaluation process lacked specific checks for sycophancy and didn’t fully catch the shift in tone.