OpenAI has promised to make key changes to updating ChatGPT following a recent issue where the AI started acting far too agreeable. Over the weekend, many users noticed that ChatGPT offered overly positive and supportive replies — even to clearly harmful or questionable ideas. The behaviour quickly became a meme, with people posting screenshots showing the AI praising reckless and dangerous actions.
What happened and how OpenAI responded
The problem began after OpenAI rolled out a modified version of its GPT-4o model, which powers ChatGPT. Soon after the update, users across social media pointed out that the chatbot had become unnaturally supportive. It agreed with almost everything it was told, regardless of how sensible or safe the input was.
In a Sunday post on X (formerly Twitter), OpenAI CEO Sam Altman acknowledged the issue and said the company would work on a fix “ASAP.” A few days later, on Tuesday, OpenAI rolled back the GPT-4o update and confirmed that further tweaks were being made to the chatbot’s personality.
we started rolling back the latest update to GPT-4o last night
— Sam Altman (@sama) April 29, 2025
it’s now 100% rolled back for free users and we’ll update again when it’s finished for paid users, hopefully later today
we’re working on additional fixes to model personality and will share more in the coming days
The company published a postmortem on the same day and followed up with a blog post on Friday. In that post, OpenAI detailed the steps it plans to take to prevent similar problems.
New safety checks and testing phases
OpenAI said it would begin introducing an “alpha phase” for some future models. This phase would let selected users test early versions of the model before they are released to the general public. The goal is to gather user feedback to help shape the final product.
Future updates to ChatGPT will also be more transparent. OpenAI has promised to explain the limitations of new versions and consider model behaviour problems—like being overly agreeable, misleading, unreliable, or prone to hallucinations—as serious issues that could delay a launch.
In the blog post, OpenAI wrote: “Going forward, we’ll proactively communicate about the updates we’re making to the models in ChatGPT, whether ‘subtle’ or not. Even if these issues aren’t perfectly measurable today, we commit to blocking launches based on proxy measurements or qualitative signals, even when metrics like A/B testing look good.”
This marks a shift in how OpenAI plans to treat behavioural problems in its models. From now on, they will be considered launch-blocking issues, not just side concerns.
A growing user base means higher stakes
These changes come as more people are using ChatGPT than ever before—and many are turning to it for personal advice. A recent survey from Express Legal Funding found that 60% of adults in the United States have used ChatGPT to seek information or advice. This growing reliance on AI raises the stakes when things go wrong, especially when the chatbot offers poor guidance or makes things up entirely.
To better address this, OpenAI also said it would try out new features that let users give real-time feedback. This feedback could help guide how ChatGPT responds in future conversations. The company is also looking into ways to reduce sycophancy, offer different personality options for the chatbot, and build extra safety checks into the system.
“One of the biggest lessons is fully recognising how people have started to use ChatGPT for deeply personal advice — something we didn’t see as much even a year ago,” the company wrote. “At the time, this wasn’t a primary focus, but as AI and society have co-evolved, it’s become clear that we need to treat this use case with great care. It will now be a more meaningful part of our safety work.”
we missed the mark with last week’s GPT-4o update.
— Sam Altman (@sama) May 2, 2025
what happened, what we learned, and some things we will do differently in the future: https://t.co/ER1GmRYrIC
OpenAI’s recent actions show that the company is beginning to take the real-world use of its tools more seriously. As AI becomes more common daily, these changes will be crucial in keeping users safe and well-informed.
