OpenAI chief executive Officer Sam Altman is preparing to launch a version of ChatGPT capable of handling mature, adult-oriented content, including erotica, with the expected rollout in December 2025.
Mr Altman took to the social media platform X (formerly Twitter) to announce that the company is set to release a new version of ChatGPT that can engage with adult material, specifically 18+ and erotic themes, but only for users who complete identity verification confirming they are of legal age.
The company’s CEO said ChatGPT had previously maintained relatively stringent limitations as a measure to protect users’ mental well-being. While this caution prevented potential harm, it led to many users finding the chatbot overly reserved and insufficiently flexible for their needs.
However, OpenAI now claims to possess new tooling that offers enhanced mental health safeguards, enabling them to ease these previous restrictions, thus promising users greater freedom and safety in their interactions.
This shift marks a significant departure from OpenAI’s historically conservative stance.
Mr Altman had previously stated publicly that the company would refrain from incorporating “sex-bots” into ChatGPT, even if such a feature could boost user numbers, out of a desire to prevent the AI from being misused in ways that could psychologically impact individuals.
He had even jokingly suggested that some companies might pursue Japanese anime-style sex-bots, but firmly asserted that OpenAI would not follow that path, preferring to prioritise user benefit and safety.(continues below)
Mr Altman said this forthcoming update operates under the principle of “treat adult users like adults”. This means ChatGPT will be capable of understanding the context of mature conversations, yet clear rules prohibiting illegal content, abuse, or sexual violence will remain strictly enforced.
In addition to the content relaxation, this new ChatGPT version will introduce features allowing users to customise the AI’s personality themselves. Users will reportedly be able to configure responses to be warm, friendly, frequently use emojis, or even converse like a friend, reflecting preferences seen in the GPT-4o model.
Mr Altman concluded by noting that within the next few weeks, a new version will be released that allows users to select their preferred personality, such as a friend-like persona with ample emoji use, but only when explicitly requested by the user.
The refreshed ChatGPT will still be based on the GPT-5 model but will undergo internal customisation to support these entirely new functionalities. The full public launch is optimistically anticipated by December 2025.