OpenAI drops plans to release an adult chatbot

OpenAI has “indefinitely” abandoned plans to release an a erotic chatbot for adults following concerns from employees and investors, the company confirmed to The Financial Times. Plans for such a feature, first announced in October 2025 for release in December last year, had already been delayed while company debated whether to release it all. It’s the second app OpenAI has decided to shelf this week, after announcing on Tuesday that it was shutting down its Sora video generator.

The adult-oriented chatbot, reportedly called “Citron mode,” is now on hold with no planned release date. The company reportedly had difficulty training models that previously avoided erotic content and also removing illegal behavior like bestiality or incest, two people familiar with the matter told the FT.

Open AI said that it wanted to conduct long-term research on the effects of erotic chats and user attachment to AI, adding that there was currently not yet enough “empirical evidence” on the subject. The company also said it wanted to focus on its core productivity tools like coding assistants and drop “side quests” like Sora and the erotic chatbot.

The idea for adult features came after OpenAI announced that it would add parental controls and automatic age detection features for ChatGPT. CEO Sam Altman said back in October that the company had always been careful about such issues over concerns around unhealthy AI attachments, but felt comfortable that it could “safely relax the restrictions in most cases.”

However, the adult mode had reportedly caused concern among investors, particularly amid the controversy caused by rival xAI’s Grok model that generated deepfake nudes of real people and children. Staff also worried about the feature, with one senior employee even leaving the company over the issue. “AI shouldn’t replace your friends or your family; you should have human connections,” he told the FT.

Another challenge is OpenAI’s age-checking tech, introduced following lawsuits from families who said that ChatGPT harmed their children. The tech reportedly has an error right higher than ten percent, which would still give a large number of young people access to the tech. OpenAI said that figure is in the industry standard range and that it is continuing to work on its accuracy.