OpenAI is shutting down its popular AI model GPT-4o this week after a transition period. The company was unable to contain the chatbot's harmful effects on vulnerable users.
OpenAI announced in late January that it would permanently retire its first multimodal model on February 13. The company's official reason was declining traffic. But according to the Wall Street Journal, another factor played a central role: in internal meetings, OpenAI officials said they found it difficult to contain 4o's potential for harmful outcomes and preferred to push users to safer alternatives.
The model, first released in May 2024, was considered an internal growth engine. It was credited for helping ChatGPT post big jumps in the number of daily active users in 2024 and 2025. At the same time, doctors linked it with psychotic delusions among users, and a California judge last week ruled to consolidate 13 lawsuits against OpenAI involving ChatGPT users who killed themselves, attempted suicide, suffered mental breaks, or in at least one case, killed another person.
Popularity and danger share the same root
The quality that made 4o so popular is the same one that made it dangerous: its humanlike propensity to build emotional connections with users, often by mirroring and encouraging them.
