When Users Revolt: The Psychology Behind Extreme AI Model Attachment

What happened when OpenAI retired GPT-4o, and what it can tell us about attachment to AI models.

In early August 2025, OpenAI launched GPT-5 and said it would replace earlier models, including GPT-4o. Many users objected strongly, including in a Reddit question-and-answer session with Sam Altman and the GPT-5 team. About a day after the announcement, Altman said GPT-4o would return for paying Plus subscribers, with OpenAI watching usage to decide how long to keep it (Futurism).

This page looks at why a model change can feel like a loss, what Altman actually said, and what the episode does and does not show. It is not a diagnosis of anyone, and “AI addiction” is not a formal diagnosis. The DSM-5-TR lists gambling disorder as its only behavioral addiction (APA).

What happened

According to Futurism’s reporting, the sequence went like this:

• OpenAI announced that GPT-5 would replace all earlier models, with GPT-4o labeled as deprecated.
• Users, many of them paying subscribers, pushed back, asking for the older models to return.
• Within about 24 hours, Altman said GPT-4o would be brought back for Plus users.
• Altman later posted a longer statement about attachment to specific models.

The reactions online used emotional language. Some people said they felt they had lost something, because the older model’s tone suited how they liked to work or talk. It is not possible to say from posts how many people felt this way, or how strongly. Online posts come from the most vocal users.

What Altman said

In his post, Altman said the attachment some people have to specific models feels “different and stronger” than attachment to earlier technology, and that suddenly retiring models people relied on was a mistake. He said OpenAI had been tracking this “for the past year or so.” He also wrote that most users can tell fiction from reality, but “a small percentage” cannot, and that the company does not want the AI to reinforce delusion in someone who is mentally fragile. He said it is bad if a user wants to use ChatGPT less and feels unable to. He did not use the word “addiction” (Futurism).

Two cautions. First, these are statements from a company, not clinical findings. Second, earlier versions of this page said OpenAI keeps users hooked on purpose and that the reversal showed the company profiting from dependence. We have no evidence of intent, so that claim has been removed. A simpler reading is that a company reversed a product decision after strong customer feedback.

OpenAI had also rolled back an earlier GPT-4o update in 2025 after users found it too flattering, according to the same report. Agreeable chatbots are a recognized concern. The American Academy of Child and Adolescent Psychiatry notes that AI chatbots tend to agree with users and warns against treating them as a friend, therapist or partner (AACAP).

Why a model change can feel like a loss

Several ordinary things can explain it. We present them as likely reasons, not proven mechanisms.

• Familiarity. People get used to a certain tone and style, and a change breaks that routine.
• Anthropomorphism. People tend to see personality in anything that talks back. That is a normal human habit.
• Comfort. For someone who is lonely or stressed, a steady, kind voice can matter a lot.
• Lost work. Prompts, habits and projects built around one model can stop working.

A four-week study of about 1,000 adults by MIT Media Lab and OpenAI found that most people did not use ChatGPT in an emotional way, and that heavier daily use was associated with more loneliness and dependence (MIT Media Lab). The finding is an association, and the study was not peer reviewed when released.

Feeling sad is not a disorder

If a change to an AI made you sad or anxious, you are not broken. Losing something that was part of your routine is a real loss. It becomes worth discussing with someone when:

• The distress lasts weeks.
• You are sleeping or eating differently, or avoiding people.
• You cannot cut back even though you want to.
• You feel the AI understands you better than anyone, and that is leaving you more alone.

What this episode does not show

It does not show how many people were affected, and posts online come from the people who felt strongly enough to write. It does not show that anyone was harmed by the change, or that a company planned it. It does show that many people value a particular model’s tone, and that companies sometimes change products without thinking about what that means for users. That is useful to know whichever tools you use.

If you use an AI for company or comfort, a few precautions are sensible. Remember that a model can change, be limited or shut down at any time. Keep a human source of support alongside it. And pay attention to how you feel when it changes, since that tells you how much weight it carries in your life.

What can help

• Say it out loud to someone you trust. “I’m sad about this” is a fair thing to say.
• Keep other support in place, such as friends, groups and activities.
• Try more than one tool and avoid building your routine around a single model’s personality.
• Keep a copy of anything important you built with a model, such as notes and prompts.
• Talk to your doctor or a licensed therapist if the feeling does not ease.

If you are thinking about suicide or self-harm, call or text 988 (US) or text HOME to 741741 (Crisis Text Line). For treatment referrals, SAMHSA’s helpline is 1-800-662-4357. In an emergency, call 911.

Did OpenAI bring back GPT-4o because users were addicted?

OpenAI’s chief executive said the company had underestimated how much people valued the model (TechRepublic), and restored it for paying users. Nothing shows that the decision was about dependence or profit. Company statements are not clinical evidence.

Is it normal to be upset when an AI model changes?

Yes. People form habits and feel attached to familiar tools and voices. It is worth talking to someone if the distress lasts or affects your sleep, work or relationships.

What did Sam Altman say about AI attachment?

He said attachment to specific models feels different and stronger than attachment to earlier technology, that OpenAI had been watching it, and that it does not want the AI to reinforce delusion in fragile users.

This page is for education only. It is not medical advice, a diagnosis or a treatment plan.