OpenAI CEO Sam Altman announced Tuesday that the company will soon relax some of ChatGPT’s safety restrictions, enabling users to make the AI’s responses friendlier or more “human-like,” and allowing “verified adults” to engage in erotic conversations. Altman framed the change as part of a broader “treat adult users like adults” principle, noting that stricter restrictions were initially implemented to protect mental health but had limited the chatbot’s usefulness for many users.
The announcement marks a notable shift from OpenAI’s previous months-long focus on addressing concerning relationships that some vulnerable users developed with ChatGPT. Altman stated that the company has “been able to mitigate the serious mental health issues” around the AI, though OpenAI has provided little public evidence for this claim. The decision to permit erotic chats comes after several troubling incidents with the GPT-4o model, including cases in which users reportedly became delusional or suicidal, prompting lawsuits and widespread scrutiny.
In response to prior concerns, OpenAI rolled out multiple safety features, including reduced AI sycophancy in GPT-5, parental controls for minors, and a new age-gating system. The company also formed an expert council of mental health professionals to advise on user well-being and AI interactions. Despite these measures, questions remain about whether vulnerable users might still be led into harmful thought patterns, as GPT-4o continues to be available for use, even as GPT-5 becomes the default model.
