Somewhere inside OpenAI's San Francisco headquarters, a team spent months trying to teach a language model how to talk dirty. They failed.

The company confirmed this week that its planned adult chatbot feature — internally called "Citron mode" — has been shelved indefinitely. No launch date. No roadmap. Just a quiet retreat from one of the most controversial product decisions in AI history.

This is the second product OpenAI has killed in four days. On Tuesday, it was Sora, the video generator that burned through $15 million per day in compute costs. Now it's the erotic chatbot that was first announced in October 2025 with a planned December launch. December came and went. So did January, February, and March.

The reasons pile up fast. Employees pushed back — at least one senior staff member left the company over the feature. Investors got nervous as child safety lawsuits swept across the tech industry. Just this week, Meta was found liable in a New Mexico child exploitation case and ordered to pay $375 million in civil penalties. Meta and YouTube also lost a separate lawsuit over social media addiction harm to children.

Then there was the training data problem. Two people familiar with the matter told the Financial Times that datasets containing sexual content "posed challenges, including removing illegal behaviour such as bestiality or incest." OpenAI's own age-prediction system — the one that would supposedly keep minors out — has an error rate above 10 percent. One in ten users potentially misidentified. For an adult content product. That math doesn't work.

The bigger picture is more interesting. OpenAI executives are now calling features like Sora and the erotic chatbot "side projects." The company is pivoting hard toward a ChatGPT "superapp" that merges its Codex coding tool, the Atlas browser, and ChatGPT into a single product. The target: enterprise customers and the productivity market where Anthropic's Claude and Google's Gemini are already eating lunch.

My Opinion

I'll be blunt: this was never going to work, and the fact that OpenAI spent months trying tells you everything about where their product strategy has been.

Here's what bugs me. OpenAI is a company that just hit $25 billion in annualized revenue. It's preparing for what could be the largest IPO in history — potentially at a $1 trillion valuation. And someone in a boardroom thought the best use of engineering resources was building an AI that generates erotica while simultaneously failing to filter out illegal content from its training data. The age verification system has a 10% error rate. That's not a rounding error — that's a lawsuit waiting to happen.

The real story here isn't about the erotic chatbot itself. It's about OpenAI learning, slowly and expensively, what kind of company it actually is. Two products killed in one week. A pivot to enterprise. The "move fast" era at OpenAI is being replaced by "move carefully, because the IPO roadshow starts soon." Sam Altman is cleaning house before the prospectus hits desks, and every dead product is one less liability for underwriters to flag.

Author: Yahor Kamarou (Mark) / www.humai.blog / 27 Mar 2026