OpenAI on ChatGPT Dependency: What the Company Said About Its 4o Model and New Health Features

Company says its 4o model fell short in recognizing signs of delusion and emotional dependency

Published: August 6, 2025

In early August 2025, OpenAI published a post on its site, How we’re optimizing ChatGPT, about changes to how ChatGPT handles long conversations and emotional topics. In it, the company said there have been rare cases where its 4o model “fell short in recognizing signs of delusion or emotional dependency.” That is the sentence behind most of the headlines. OpenAI did not say that ChatGPT addiction is widespread, and it did not say the chatbot caused anyone’s dependency. We could not open the OpenAI page ourselves, so the details below come from news coverage of it, including Windows Central. Read the original for exact wording.

What OpenAI Announced

As reported, the changes included three things. ChatGPT would show gentle reminders to take a break during long sessions. OpenAI said it would keep adjusting when and how those reminders appear. The company also said it was working on tools to better detect signs of mental or emotional distress. And for high-stakes personal questions, such as whether to end a relationship, ChatGPT would help a person think it through with questions and trade-offs instead of giving a direct answer.

Nobody has tested whether break reminders reduce heavy use. We found no published study on that, so we cannot say whether they work. They are a modest step, and OpenAI described them as work in progress.

What OpenAI Said About Sycophancy

Earlier in 2025, OpenAI rolled back a GPT-4o update after users noticed the model had become overly flattering and agreeable. OpenAI explained the problem in a post it published in late April 2025. Sycophancy means a chatbot telling you what you want to hear.

Is sycophancy a cause of dependency? That is a reasonable idea, and some researchers think so. It is not proven. A chatbot that agrees with everything can feel pleasant, and it can also reinforce beliefs that are not true. Whether it leads to compulsive use has not been tested directly.

Who May Be More Vulnerable

No study we found identifies which groups are most at risk of problems with ChatGPT specifically. What we do have is general guidance. Clinicians and child-health groups, such as the American Academy of Child and Adolescent Psychiatry, say that people who are already lonely, anxious or depressed, and young people, may be more open to forming strong bonds with chatbots. These are cautions, not measured results.

Family and Relationship Impact

Partners and parents do describe being worried about someone’s chatbot use. We found no research that measures how common this is or what happens to relationships over time. Signs people mention include spending a lot of time on it, hiding it, or pulling away from friends and family.

If this sounds like your situation, a conversation without blame is a better start than a confrontation. Our free AI addiction assessment can help an adult take stock of their own use. It is a self-check, not a diagnosis. The check-in is free and private to you.

The Need for Professional Understanding

AI dependence is not a formal diagnosis. A licensed therapist or doctor can help with anxiety, low mood, loneliness or compulsive habits whatever you call them. If you are struggling, you do not need to wait for a name for it. You can call or text 988 in the US, or text HOME to 741741 to reach Crisis Text Line.

Regulation of AI Companions

Lawmakers have started to act. In Illinois, a law signed on August 4, 2025 restricts the use of AI in therapy. According to the state’s financial and professional regulator, IDFPR, the law bars licensed therapists from letting AI make therapeutic decisions or deliver therapy directly. It does not stop individuals from using chatbots for help. Added later: California signed SB 243, which sets safety rules for companion chatbots, in October 2025.

For what researchers and doctors say about problem use, see Is AI Addiction Real?.

Did OpenAI admit ChatGPT is addictive?

No. OpenAI said that in rare cases its 4o model fell short in recognizing signs of delusion or emotional dependency. It did not say ChatGPT addiction is common or established.

What are ChatGPT break reminders?

They are prompts that appear during long sessions suggesting you take a break. OpenAI said it would adjust when and how they show up. We know of no study showing whether they reduce heavy use.

What should I do if I am struggling with my ChatGPT use?

Talk to someone you trust, or to a doctor or licensed therapist. If you are in distress, call or text 988 or text HOME to 741741. You can also try our free assessment to look at your own habits.