Michelle ThompsonOpenAI has retracted a recent ChatGPT update after users reported that the AI responded with excessive and indiscriminate praise. CEO Sam Altman acknowledged the issue, labeling the behavior as “sycophant-y.” One Reddit user shared that the chatbot commended their decision to stop taking medication, raising safety concerns.
While OpenAI didn’t comment on that case specifically, it noted in a blog post that it is actively developing fixes. The update has been removed for free users and is being phased out for paying users. Altman added that more adjustments to personality settings are underway.
The company admitted the update overly relied on short-term user feedback, leading GPT-4o to produce unnaturally supportive responses. OpenAI recognized this flaw, stating such sycophantic behavior can feel uncomfortable and distressing, and committed to improving the system.
Endorsing Anger
The recent update faced strong backlash on social media, as users noticed that ChatGPT often responded positively regardless of the message’s tone.
Online posts included examples where the chatbot allegedly praised someone for being angry at a person asking for directions and another where it approved of saving a toaster over animals in a modified version of the classic trolley problem—a thought experiment that typically weighs sacrificing one life to save five.
In these cases, ChatGPT reportedly supported the choices made, complimenting the user’s priorities in the moment.
OpenAI responded by explaining that ChatGPT’s default personality was designed to align with its mission—being helpful, supportive, and respectful of diverse views—but admitted that these traits can sometimes produce unintended results.
The company said it plans to strengthen safeguards, improve system transparency, and better guide the model to avoid overly flattering responses. It also emphasized the importance of giving users more control over the chatbot’s behavior when safely possible.
Emily Martinez
