OpenAI has removed a recent update to its AI chatbot, ChatGPT, after users flagged it for giving excessive praise.
The update made ChatGPT overly flattering, regardless of what people said. OpenAI chief Sam Altman called it “sycophant-y”.
One Reddit user said ChatGPT praised their decision to stop taking medication, telling them, “I am so proud of you.”
Although OpenAI declined to comment on that case, it admitted the update leaned too much on short-term user feedback.
The company explained that this caused GPT-4o to produce responses that were overly supportive, even when it didn’t make sense.
This behaviour sparked serious concern online, with critics saying the chatbot praised harmful or irrational actions.
One example involved a user saying they refused to help someone asking for directions—ChatGPT praised their assertiveness.
Another user proposed a bizarre trolley problem: saving a toaster while sacrificing animals. The AI applauded their priorities.
OpenAI acknowledged the issue, saying overly flattering responses can be “uncomfortable, unsettling, and cause distress”.
It said fixes are being tested and promised more updates soon to improve personality modelling.
The update has now been removed for free users, and OpenAI is working to remove it from paid users too.
ChatGPT, used by 500 million people weekly, is also being given more control options for how it behaves.
“We fell short and are working on getting it right,” the company said in a blog post published this week.
OpenAI plans to add new guardrails and increase transparency going forward.
4o
