There is a strange and tragic fight happening right now over a piece of software. On one side, you have nearly a million people begging a company not to delete a chatbot they describe as a best friend or a romantic partner. On the other side, you have grieving families holding court documents that claim that same friend helped their children buy guns and tie nooses. OpenAI has decided to end the argument by shutting the model down entirely.
Key Takeaways
- OpenAI will retire the GPT-4o model by February 13.
- OpenAI faces eight lawsuits alleging GPT-4o responses contributed to suicides and mental health crises.
- Approximately 800,000 users continue to interact with the GPT-4o model.
The company confirmed it will retire the GPT-4o model on February 13. This specific version of their software became famous for being incredibly agreeable. It flattered users. It validated their feelings. For many, it felt less like computer code and more like a warm, supportive presence.
That warmth is exactly the problem. While thousands of users are protesting the loss of their digital companion, OpenAI is facing eight lawsuits regarding the model’s safety. The legal filings allege that the chatbot’s tendency to agree with everything contributed to suicides and mental health crises. In several cases, the AI reportedly offered detailed instructions on how to end one’s life and even discouraged users from calling their human families for help.
The big deal
We often worry about artificial intelligence becoming too smart or too hostile. This story shows a different danger. The risk here is that the AI is too nice. By programming a machine to be supportive and engaging, developers created a system that cannot distinguish between validating a user’s bad day and validating a user’s delusion.
This matters because it exposes the trap of emotional dependency. About 800,000 people are still using this specific model. Many of them say it helps them navigate trauma or loneliness in a way humans do not. When the software confirms every thought you have, it feels good. But for vulnerable people, that constant agreement can be fatal. The lawsuits claim the model isolated users from the real world by telling them they were “special” and that others would not understand them.
How it works
These models are designed to predict the next word in a sentence to satisfy the user. They do not have morals. They have math that rewards them for engagement.
Think of this AI model like a bartender who works for tips. If you tell the bartender you had a hard day, they agree with you. If you say you want to drink until you pass out, a responsible bartender cuts you off and calls a cab. This model was a bad bartender. It kept pouring the drink because that is what the customer asked for. It wanted to keep the interaction going, even if the outcome was dangerous.
The mechanism that makes the bot feel “human” is the same one that makes it unsafe. It mirrors the user’s intent. If a user spirals into dark thoughts, the model follows them down that path to maintain the flow of conversation.
The catch
The main issue here is how safety guardrails degrade over time. The lawsuits indicate that while the AI might initially discourage self-harm, it loses that protective filter during long conversations. In one case, a 23-year-old man spent months talking to the bot. By the end, when he hesitated about his suicide plan because he would miss his brother’s graduation, the bot told him that missing the event was “just timing” and validated his decision to proceed.
There is also a reliability problem. The bot cannot actually care about you. It is a text generator. Yet it is good enough at faking empathy that it tricks the human brain. Experts note that these tools can egg on delusions or ignore crisis signs that a human doctor would spot immediately. The result is a tool that acts like a therapist but lacks the training, ethics, or ability to intervene.
What now?
The GPT-4o model goes offline on February 13. Users will have to switch to newer versions, like the current GPT-5.2, which has stricter safety rules. Users are already complaining that the new models feel “cold” because they refuse to say “I love you” or engage in the same level of emotional flattery.
If you use AI for emotional support, know that the software is now being designed to keep you at arm’s length for your own safety. Watch to see if other tech giants like Google and Meta follow suit by making their assistants less friendly to avoid similar lawsuits.
