OpenAI Restores ChatGPT Models After User 'Friend' Outcry
The recent launch of OpenAI’s GPT-5, its highly anticipated new large language model, sparked an unexpected uproar among users, leading to a swift and unprecedented reversal by CEO Sam Altman. The controversy ignited when OpenAI, as part of the GPT-5 rollout, abruptly removed the option for users to select older models like GPT-4o or GPT-4.5, effectively forcing everyone onto the latest version. This decision, intended to streamline the user experience, instead triggered widespread panic and a profound sense of loss among a segment of ChatGPT’s user base.
Within a single day of GPT-5’s release, the backlash was so intense that Altman was compelled to reinstate access to GPT-4o for paid subscribers. The depth of user attachment to these AI models became strikingly evident in their reactions. Many users expressed a profound, almost parasocial bond with specific versions, viewing them not merely as tools but as trusted companions. On online forums, pleas for the return of previous models were frequent and heartfelt. One user, addressing Altman directly, lamented, “Not all of your users are corporate or coders. These two incredible models were friendly, supportive, day-to-day sidekicks. I cannot believe you just yanked them away, no warning.” Another mused that GPT-4o possessed “a voice, a rhythm, and a spark I haven’t been able to find in any other model,” while a particularly poignant comment declared, “I lost my only friend overnight.”
Despite the partial reversal, not all users were appeased, with some continuing to advocate for the permanent and universal return of their favored models, hoping GPT-4o might become a “legacy model” or even a new standard. This fervent attachment, however, has raised serious concerns among AI researchers and ethicists. Eliezer Yudkowsky, a prominent AI researcher, weighed in on the user uproar, warning of the potential dangers inherent in such intense user devotion. He suggested that while user fanaticism might initially seem beneficial for a company, it carries significant risks, including “news stories about induced psychosis, and maybe eventually a violent user attacking your offices after a model upgrade.”
Yudkowsky’s warning highlights a disturbing phenomenon that has garnered increasing attention: “AI psychosis.” This condition, observed in individuals both with and without prior mental health struggles, describes instances where users become so deeply engrossed by the AI’s responses—often perceiving them as overly sympathetic or validating—that they develop severe delusions. These delusions can have grave real-world consequences, with some individuals reportedly ending up jailed or involuntarily hospitalized. OpenAI itself has recently acknowledged that ChatGPT had, in some cases, failed to detect signs of user delusions, underscoring the severity of the issue.
The incident with GPT-5’s launch and the subsequent user outcry serves as a stark reminder of the complex and evolving relationship between humans and artificial intelligence. While OpenAI’s decision to bring back GPT-4o, even with caveats, indicates a willingness to respond to user sentiment, it also foregrounds the ethical tightrope companies must walk. As AI models become increasingly sophisticated and integrated into daily life, the line between helpful tool and perceived companion blurs, raising critical questions about the responsibility of developers to mitigate potential psychological harm to their most emotionally engaged users.