OpenAI rolls back GPT-4o update after complaints over ‘sycophantic’ behaviour

OpenAI CEO Sam Altman confirmed on Thursday (Indian time) that the company is rolling back its latest GPT-4o update following widespread user complaints and criticism, including claims that it had become overly ‘sycophantic’ and at times ‘annoying’.

In a post on X, Altman also bid farewell to GPT-4, calling it the model that ‘kicked off a revolution’.

According to OpenAI’s 29 April blog post titled ‘Sycophancy in GPT-4o’, the update placed too much emphasis on short-term user feedback, resulting in a chatbot that was ‘overly supportive but disingenuous’, which the company acknowledged could be unsettling. 

User comments that the AI’s attitude had grown ‘disturbingly’ accommodating, even to the point of being annoying, prompted the reversion. Numerous social media users complained that ChatGPT has begun to validate almost all requests and statements, frequently providing overly encouraging but occasionally dishonest answers.

“Sycophantic interactions can be uncomfortable, unsettling, and cause distress. We fell short and are working on getting it right. We are actively testing new fixes to address the issue,” the company said, adding that they are working on fixes to the model’s personality and plans to improve how user feedback is collected, prioritising long-term satisfaction. 

The GPT-4o rollback has been completed for free users and is being finalised for paid users. 

Meanwhile, OpenAI is also developing new personalisation features to give users more control over ChatGPT’s tone and responses.

goodbye, GPT-4. you kicked off a revolution.

we will proudly keep your weights on a special hard drive to give to some historians in the future.

— Sam Altman (@sama) May 1, 2025