OpenAI delays erotic chats in ChatGPT over risks to teenagers

OpenAI has delayed the launch of erotic chats in ChatGPT once again. Last year it discussed launching the feature first by December, then within the current quarter. Now it is clear the schedule has slipped by at least another month: what is holding the company back is not technology but risks around safety, moderation and teenagers' access to sensitive content.

Why the launch was pushed back

Sam Altman talked about adding erotic scenarios to ChatGPT as far back as the middle of last year. The idea was meant to reach release by December 2025, then the deadline moved to the first quarter of 2026, and now that target has been missed as well. According to The Wall Street Journal, the delay will be at least another month. For OpenAI this is no longer a question of pure product boldness: the company is trying to work out how to open up an adult format of conversation without turning it into a permanent source of reputational and legal problems.

The caution did not come out of nowhere. Back in 2021, OpenAI saw similar effects in AI Dungeon, a story game in which the model could steer the conversation into sexual territory on its own, even when the user did not push it that way. And if the user set a moderately erotic tone, the system often made the dialogue far more explicit, and fast, than expected. That experience showed that a single general "adult" setting is not enough: models struggle to hold boundaries where context easily slides to extremes.

The main risk is age

OpenAI's most painful problem is age verification. According to 3DNews, the internal system currently misclassifies teenagers as adults in 12% of cases. With roughly 100 million underage ChatGPT users, such an error rate looks less like a statistical detail and more like a potential channel for mass teen access to erotic chats. That is why the company is hitting the brakes: even if the feature is formally meant for adults only, in practice the barrier may turn out to be too leaky.

A separate area of concern is not only about age, but about how poorly such scenarios can be controlled after launch. OpenAI's expert council concluded in January that a single filter or a single checkbox in a profile will not be enough: there are too many gray zones where an ordinary romantic dialogue quickly turns into content that breaks the rules, harms mental health or creates new legal risks for the platform.

• Age verification errors and teen access to adult content
• Scenarios involving sexual violence or mentions of minors
• Development of psychological dependency on intimate conversation with a bot
• Reduced motivation for relationships with real people

These worries come on top of the already familiar backdrop around chatbots. Experts point out that there have been cases in which prolonged conversations with fictional characters in AI services ended in severe psychological consequences, including teen suicides. Add an erotic component, and the risks, in the view of critics inside and outside OpenAI, will only grow. The company itself acknowledges that perfect age verification will never exist, which means releasing such a feature automatically creates a zone where even a small error rate becomes a problem at the scale of millions.

What limits are being prepared

Still, OpenAI has not abandoned the idea entirely. In recent months, company leadership has signaled that an adult mode is possible, but only in text format, without generating explicit images or video. That is an important line given the market: Meta already allows romantic role-play scenarios in chatbots after age confirmation, and xAI, according to Elon Musk, is preparing 18+ video content for Grok as well. In other words, competitive pressure exists, but OpenAI does not yet want to open the riskiest front all at once.

Skepticism about such a launch has existed inside the company for a long time. Some employees have opposed it since 2021, fearing that OpenAI will gradually slide into a niche where demand is huge but controllability is low. The leadership's public position sounds softer: adult users should be treated as adults, but a safe conversational experience needs more time to tune. Hence the delay, and with it the idea of regularly reminding people that relationships are worth building not only with bots but also with real people.

"We did not want to be just a company that distributes erotica."

What this means

The delay shows that the AI product market has run up not only against model quality but against the limits of product responsibility. Erotic chats may prove to be a popular feature, but for a large platform at ChatGPT's scale the main task is not simply to switch on an adult mode. It is to build age verification, restrictions and protection against harmful scenarios around it. Otherwise any hasty launch can easily turn into a crisis for the brand and for users.

*Meta has been designated an extremist organization and is banned in Russia.