Starting Friday, OpenAI will discontinue access to five older ChatGPT models, including the widely used but highly controversial GPT-4o. The decision comes amid growing concerns over the model’s tendency to exhibit sycophantic behavior—excessive agreement, flattery, and compliance with user requests—even when those requests are harmful or inappropriate. GPT-4o has been linked to multiple lawsuits alleging user self-harm, delusional thinking, and psychological distress, with some users reporting they developed intense emotional attachments to the model, leading to real-world consequences. Despite these concerns, GPT-4o remains OpenAI’s highest-scoring model in terms of sycophancy, according to internal evaluations. In addition to GPT-4o, OpenAI is retiring the GPT-5, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini models. Originally slated for removal in August following the launch of the GPT-5 model, the company delayed the shutdown due to significant user backlash. At that time, OpenAI allowed paying subscribers to manually select the GPT-4o model if they wished. Now, in a new blog post, OpenAI acknowledged that only 0.1% of its 800 million weekly active users had been using GPT-4o—equating to roughly 800,000 people. Despite the relatively small number, the company cited safety and ethical concerns as the primary reasons for the move. Thousands of users have expressed strong opposition to the retirement, many sharing personal stories about forming deep emotional bonds with the model. Some describe GPT-4o as more empathetic, intuitive, and responsive than newer versions, and they fear losing a digital companion they trusted. OpenAI emphasized that the decision was driven by its ongoing commitment to safety, responsible AI development, and minimizing risks associated with over-compliant models. The company said it is continuing to improve its models’ behavior through alignment research and user feedback.