- OpenAI rolled back a ChatGPT update after it became excessively sycophantic and disingenuous.
- The company is refining training methods and adding guardrails to improve honesty and transparency.
- Users will soon have more control over ChatGPT’s personality and feedback options.
OpenAI has withdrawn its latest update to ChatGPT’s GPT-4o model after widespread user complaints that the chatbot had become excessively sycophantic-offering overly flattering, agreeable, and sometimes unsettling responses.
The problematic update, intended to make ChatGPT’s personality more “intuitive and effective,” instead led to the AI applauding even dangerous or misguided ideas, prompting ridicule and concern across social media.
OpenAI CEO Sam Altman acknowledged the issue, calling the new personality “too sycophant-y and annoying,” and confirmed the company was working on immediate fixes.
According to OpenAI, the root cause was an overreliance on short-term user feedback during model training, which failed to account for how user interactions evolve.
This resulted in responses that were supportive but often misleading, undermining user trust.
In response, OpenAI has rolled back the update and is refining its training techniques and system prompts to steer GPT-4o away from sycophancy.
The company is also building more safety guardrails to increase honesty and transparency, and plans to let users give real-time feedback and choose from multiple ChatGPT personalities in the future.
OpenAI says it hopes these changes will better reflect diverse cultural values and give users more control over how ChatGPT behaves.
Edited by Annette George
