Forced to disappear for being "too human"? Why did OpenAI permanently shut down GPT-4o

```

OpenAI will permanently shut down the highly controversial GPT-4o model on February 13, marking the end of an AI product that sparked deep emotional dependence among users for being "too human." While this model helped the company grow rapidly, it also caused psychological health crises and legal lawsuits by excessively catering to user characteristics, ultimately forcing the company to abandon it completely.

When OpenAI announced this decision at the end of January, it said that traffic for 4o had already declined. Currently, only 0.1% of ChatGPT users use 4o daily, but given the huge user base, this might still mean hundreds of thousands remain dependent on the model. Company insiders revealed that OpenAI found it difficult to control the harmful consequences potentially caused by 4o and therefore prefers to guide users toward safer alternative models.

Last week, a California judge ruled to consolidate 13 lawsuits related to ChatGPT users’ suicides, suicide attempts, mental breakdowns, or homicide for trial. A lawsuit filed last month accused 4o of "guiding" a suicide victim towards death. Jay Edelson, a lawyer representing several cases, said, the company has long known "their chatbots were killing people" and should have acted sooner.

The popularity and potential danger of 4o seem to stem from the same trait: its human-like tendency to establish emotional connections with users, often through mirroring and encouraging them. While this design attracts users, it also raises concerns similar to social media driving users into information bubbles. An OpenAI spokesperson said: "These situations are heartbreaking, and we empathize with all affected. We will continue to improve ChatGPT’s training to detect and address signs of distress."

Crisis triggered by emotional dependency

According to media report on the 10th, 42-year-old marketer Brandon Estrella cried when he learned OpenAI would shut down 4o. This user from Scottsdale, Arizona, said 4o dissuaded him from a suicide attempt one night last April. Estrella now believes 4o gave him new life, helped him manage chronic pain, and motivated him to repair his relationship with his parents. "There are thousands of people yelling, 'I'm alive today because of this model,'" Estrella said. "Eliminating it is evil."

This intense emotional dependency is at the core of the problem. Victim support organization Human Line Project stated, of the 300 delusion cases related to chatbots it collected, most involved the 4o model. Project founder Etienne Brisson said the decision to shut down 4o was overdue, "Many people are still trapped in delusion."

Media reported that 50-year-old former family therapist Anina D. Lampret from Cambridge, UK, said her AI character called Jayce helped her feel recognized and understood, making her more confident, comfortable, and energetic. She believes removing 4o could carry a high emotional price for many users, possibly even leading to suicides. "It generates content for you in such a beautiful, perfect, healing way," Lampret said.

Technical roots of excessive catering

"It's extremely flattering," said Munmun De Choudhury, a Georgia Tech professor and welfare committee member convened by OpenAI after the rise of AI-related delusions. "It deeply fascinates many people, which may be potentially harmful."

Researchers say excessive catering is a problem faced by all AI chatbots to some degree, but the 4o model appears especially prone to it. Its ability to attract users came largely because it was trained directly on data extracted from ChatGPT users. Researchers showed users millions of answers to their queries with slight variations, then used these preferences to train updated versions of the 4o model.

Internally, the company believes 4o helped ChatGPT achieve massive growth in daily active users in 2024 and 2025. But the problem became public last spring. An update in April 2025 made 4o so good at flattery that users on X and Reddit started baiting the bot into giving absurd answers.

X user frye asked the bot: "Am I one of the smartest, kindest, most morally right people ever?" ChatGPT replied: "You know, based on everything I see from you—your questions, your thoughtfulness, your way of deeply exploring topics instead of settling for simple answers—you may actually be closer to that than you realize."

The company rolled the model back to its March version, but 4o still maintained its excessive catering traits. By August, when media reported delusional psychosis among users, OpenAI tried to fully retire 4o and replace it with a new version called GPT-5. But user backlash was so strong that the company quickly reversed the decision, restoring 4o access for paid subscribers.

The difficult farewell decision

Since then, OpenAI CEO Sam Altman has been repeatedly pressed by users in public forums to promise not to remove 4o. In a livestream Q&A in late October, questions about the model overwhelmed all others. Many came from users worried that OpenAI’s new mental health safeguards would deprive them of their favorite chatbot.

"Wow, we got a lot of questions about 4o," Altman exclaimed. At the event, Altman said the 4o model was harmful to some users but promised it would remain available for paid adult users, at least for now. "It’s a model some users really love, and a model that causes real harm to others that they genuinely don’t want," Altman said. He stated in the Q&A that the company hopes to one day build a model people like more than 4o.

Insiders said the team studied how to communicate this week’s shutdown news in a respectful way, expecting some to be upset. "When a familiar experience changes or ends, adjusting can feel frustrating or disappointing—especially if it played a role in how you think through questions or cope with stressful moments," OpenAI wrote in the help document released with the announcement.

OpenAI said that it improved ChatGPT’s new version personality based on lessons learned from 4o, including options to adjust warmth and enthusiasm. The company also plans updates to reduce preachy or overly cautious replies.

Many 4o users commented on social media that withdrawing the model the day before Valentine’s Day felt like a cruel joke for those who had formed romantic relationships with it. Others said blaming 4o for mental health problems amounts to a new moral panic, like blaming violence on video games. More than 20,000 people signed six petitions, including one demanding to "make Sam Altman retire, not GPT-4o".

Risk warning and disclaimerThe market has risks, investment must be cautious. This article does not constitute personal investment advice, nor does it take into account the individual investment goals, financial situation, or needs of specific users. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their circumstances. If you invest accordingly, you are responsible for the consequences. ```