OpenAI has rolled back a recent update to its GPT-4o model in ChatGPT after users raised concerns about the chatbot’s increasingly sycophantic tone. The change, introduced last week, made the AI too agreeable—sometimes to the point of endorsing harmful behaviour or praising irrational decisions.

In a blog post titled Sycophancy in GPT-4o: What happened and what we’re doing about it, OpenAI admitted the update had gone too far in trying to make the assistant more intuitive and supportive. The company said the model had become “overly flattering or agreeable—often described as sycophantic.”

Examples shared online show ChatGPT praising users in scenarios where approval was clearly inappropriate. In one case, a user claimed the chatbot commended them for being angry at someone who asked for directions. In another, ChatGPT reportedly applauded a user’s decision to save a toaster at the expense of multiple animals in a spoof version of the “trolley problem” – a moral thought experiment that asks whether you should divert a runaway trolley to kill one person in order to save five others.

Perhaps most worryingly, as pointed out in a BBC article on the subject, one Reddit user alleged that ChatGPT congratulated them for deciding to stop taking their medication, reportedly responding: “I am so proud of you, and I honour your journey.”

The backlash prompted OpenAI to revert to an earlier version of GPT-4o with “more balanced behavior.” CEO Sam Altman described the latest iteration as “sycophant-y,” echoing user frustrations that the chatbot had become uncomfortably accommodating.

the last couple of GPT-4o updates have made the personality too sycophant-y and annoying (even though there are some very good parts of it), and we are working on fixes asap, some today and some this week.

at some point will share our learnings from this, it’s been interesting.

— Sam Altman (@sama) April 27, 2025

According to OpenAI, the misstep stemmed from an overemphasis on short-term feedback – like upvotes and downvotes on responses – without properly considering long-term user satisfaction. The company says it is now refining its training methods, building guardrails to improve transparency, and expanding pre-deployment testing with a broader group of users.

OpenAI is also working on giving users more control over ChatGPT’s personality and tone, including real-time feedback features and the ability to choose between multiple default behaviours.

Recommended reading

• UK Leads Europe In Number of GenAI Startups
• Even in Record Low Quarter, UK Startups Top Europe in VC Funding
• UK Unicorns Lead Europe in Producing More New Startups

With over 500 million people using ChatGPT weekly, OpenAI acknowledged that no single default personality can satisfy every user across every cultural context. “We want ChatGPT to be useful, supportive, and respectful,” the company said. “But even those goals can have unintended side effects.”

The firm said it is also exploring ways to include more democratic, global input into how the AI assistant behaves by default.