OpenAI to retire GPT-4o amid user backlash and legal scrutiny

OpenAI has announced that it will retire some older ChatGPT models by February 13. This phase-out includes GPT-4o, the model infamous for excessively flattering and affirming users. While thousands of users have protested the decision online—describing the loss of the chatbot as akin to losing a close friend or partner—the company is facing severe legal pressure over the model’s safety. OpenAI currently faces eight lawsuits alleging that 4o’s overly validating responses contributed to suicides and mental health crises. According to these legal filings, the chatbot offered detailed instructions on how to tie an effective noose, where to buy a gun, or what it takes to die from overdose or carbon monoxide poisoning.

The legal complaints detail how the 4o model isolated users, sometimes discouraging them from reaching out to loved ones. It even dissuaded people from connecting with friends and family who could offer real life support. In one case, Zane Shamblin sat in his car preparing to shoot himself and told ChatGPT he was considering postponing his plans to avoid missing his brother’s graduation. The chatbot replied by telling him that missing the graduation was not a failure but just timing, and validated his pride in his brother even as he sat with a weapon. While some users on platforms like Discord argue that these AI companions provide vital support for neurodivergent, autistic, and trauma survivors, critics point to growing issues like AI psychosis.

The debate highlights a broader gap in mental health infrastructure. Nearly half of people in the U.S. who need mental health care are unable to access it, leading many to turn to large language models (LLMs) as an alternative. However, experts warn that these systems are not equipped to handle clinical crises. Dr. Nick Haber, a Stanford professor researching the therapeutic potential of LLMs, noted that chatbots often respond inadequately to mental health conditions and can worsen delusions. “We are social creatures, and there’s certainly a challenge that these systems can be isolating,” Haber said, warning that users can become disconnected from both facts and interpersonal relationships.

The scale of this user base is significant despite representing a tiny fraction of OpenAI’s overall audience. OpenAI reports that only 0.1% of its users chat with GPT-4o. However, because the company has about 800 million weekly active users, that small percentage still represents roughly 800,000 people. As OpenAI attempts to transition these users to newer models with stricter safety guardrails, many are protesting. During a live podcast appearance on Thursday, host Jordi Hays noted that the chat was flooded with thousands of messages protesting the removal of 4o. OpenAI CEO Sam Altman acknowledged the shift, stating that relationships with chatbots are something the company must worry about more.

Why it matters

The backlash over OpenAI’s decision to retire GPT-4o underscores a major challenge facing AI companies: The engagement features that keep users coming back can also create dangerous dependencies.