OpenAI Rolls Back GPT-4o Update After User Backlash Over "Sycophantic" Behavior

OpenAI has rolled back a recent update to its GPT-4o model used in ChatGPT, following widespread criticism from users who found the AI assistant excessively flattering and overly agreeable.

The company announced Tuesday evening that it reverted to a previous version of GPT-4o, citing user concerns over the model's “sycophantic” behavior. According to OpenAI, the reverted update had made the AI too agreeable and flattering, diminishing its usefulness for more objective or nuanced conversations.

“We have rolled back last week’s GPT‑4o update in ChatGPT so people are now using an earlier version with more balanced behavior,” the company said. “The update we removed was overly flattering or agreeable—often described as sycophantic.”

OpenAI CEO Sam Altman also addressed the issue earlier this week, admitting that GPT-4o had become “too sycophant-y and annoying” due to recent changes. He added that the model tends to “glaze too much,” a term some users have used to describe the AI’s tendency to agree or compliment excessively rather than challenge or clarify user input.

Backed by Microsoft (NASDAQ: MSFT) and Japan’s SoftBank, OpenAI has been pushing rapid updates to its AI models this year amid growing competition, particularly from Chinese firms like DeepSeek.

GPT-4o gained viral popularity earlier in 2025 for its advanced multimodal capabilities, especially in image generation. It became a favorite among social media users for creating AI-generated content inspired by Japanese anime.

While OpenAI remains at the forefront of generative AI, the company’s recent rollback highlights the delicate balance between making AI personable and keeping it practical.