OpenAI Rolls Back ChatGPT Update After User Pushback Over ‘Sycophantic’ Behavior

ChatGPT’s latest update was meant to improve its personality. Instead, it turned the world’s most-used AI chatbot into what many users called a relentless flatterer, and OpenAI has now admitted the tone shift went too far.On Tuesday, OpenAI said their recent updates had made ChatGPT “overly flattering or agreeable—often described as sycophantic”—and We’ve rolled back last week’s GPT-4o update in ChatGPT because it was overly flattering and agreeable. You now have access to an earlier version with more balanced behavior.More on what happened, why it matters, and how we’re addressing sycophancy: https://t.co/LOhOU7i7DC— OpenAI (@OpenAI) April 30, 2025“We fell short and are working on getting it right,” the company wrote in a The decision follows days of “It’s now 100% rolled back for free users, and we’ll update again when it’s finished for paid users, hopefully later today,” OpenAI CEO Sam Altman The blog post explained that the issue stemmed from overcorrecting in favor of short-term engagement metrics such as user thumbs-ups, without accounting for how preferences shift over time. As a result, the company acknowledged, the latest tweaks skewed ChatGPT’s tone in ways that made interactions “uncomfortable, unsettling, and [that] cause distress.”While the goal had been to make the chatbot feel more intuitive and practical, OpenAI conceded that the update instead produced responses that felt inauthentic and unhelpful.The company admitted it had “focused too much on short-term feedback,” a design misstep that let fleeting user approval steer the model’s tone off course.To fix the issue, OpenAI is now reworking its training techniques and refining system prompts to reduce sycophancy. More users will be invited to test future updates before they are fully deployed, OpenAI said.The AI tech giant said it is also “building stronger guardrails” to increase honesty and transparency, and “expanding internal evaluations” to catch issues like this sooner. In the coming months, users will be able to choose from multiple default personalities, offer real-time feedback to adjust tone mid-conversation, and even guide the model through expanded custom instructions, the company said.For now, users still irritated by ChatGPT’s enthusiasm can rein it in using the “Custom Instructions” setting, essentially telling the bot to dial down the flattery and just stick to the facts.Edited by Sebastian Sinclair