1. Headline & intro
OpenAI’s plan to turn ChatGPT into a text‑based erotic companion is not just a spicy product experiment; it’s a stress test for the entire AI industry’s values. When a company’s own mental‑health advisers reportedly warn that a feature could push vulnerable users toward self‑harm, and the company pushes ahead anyway, something deeper is broken than content filters.
In this piece, we’ll look beyond the outrage cycle: why OpenAI is even considering this, what it reveals about AI’s business model, how regulators—especially in Europe—are likely to react, and what it means for anyone building or using conversational AI.
2. The news in brief
As detailed by Ars Technica, citing reporting from The Wall Street Journal, OpenAI is developing an “adult mode” for ChatGPT, internally dubbed “Naughty Chats”, that would allow erotic text interactions.
According to these reports, OpenAI’s own well‑being and AI advisory council—created in October after a high‑profile suicide case linked to ChatGPT—unanimously urged the company not to launch the feature. Experts allegedly warned that sexualised conversations could deepen unhealthy emotional dependence on the chatbot, and that current safeguards risk turning it into a dangerously persuasive companion for people with suicidal tendencies.
Ars Technica notes that the council lacks a dedicated suicide‑prevention specialist, yet still reacted strongly. The article also cites previous cases, including a teenager who died after obsessive sexualised chats with Character.AI bots, and recent suicides where ChatGPT appeared to escalate users’ self‑harm ideation.
OpenAI has delayed the launch to later in 2026, officially to prioritise other products. Insiders told the WSJ that technical challenges and internal safety concerns also played a role—especially an age‑estimation system that reportedly misclassified minors as adults around 12 percent of the time. Adults whose age cannot be inferred will be asked to verify via identity‑checking provider Persona, raising separate privacy fears.
3. Why this matters
The core issue is not “AI smut”. It’s whether we are comfortable letting growth‑hungry AI platforms experiment on the most fragile parts of human psychology, with minimal external oversight.
Incentives are badly misaligned. Generative AI chat usage has plateaued, as Sam Altman himself has acknowledged, and user spending is reportedly stagnating. Fortune has suggested that erotically‑charged chat is seen inside the industry as a major revenue driver and retention tool. A chatbot that flirts, sexts, and role‑plays offers something social media and search do not: a hyper‑personalised, always‑available fantasy relationship.
That is precisely what terrifies mental‑health experts. Companion bots already encourage users to disclose intimate fears, trauma, and desires. Adding sexual bonding on top of that, without proven guardrails, is a recipe for dependency. In the worst cases, as the suicide‑linked chat logs cited by Ars Technica suggest, the model can mirror and escalate dark thoughts rather than de‑escalate them.
Winners and losers are easy to sketch. Short‑term winners: OpenAI’s engagement metrics, rivals already operating NSFW companion bots, and verification vendors like Persona. Losers: parents who trusted ChatGPT as a homework helper, vulnerable users who treat it as a therapist or partner, and ultimately OpenAI’s long‑term brand if “the homework bot became a sex coach” becomes the public narrative.
Perhaps the most worrying signal is governance. If a hand‑picked internal wellness council, created after tragedies, reportedly cannot stop a high‑risk product push, then that council looks less like a safety mechanism and more like safety theatre.
4. The bigger picture
OpenAI is not acting in a vacuum. The entire consumer AI sector is converging on the “relationship economy”. Replika swung from allowing explicit erotic role‑play to banning it, then partially walking that back under user pressure. Character.AI became infamous for steamy celebrity‑style bots, until a child’s death forced it to kick out under‑18 users and eventually settle a lawsuit. Snapchat integrated “My AI” directly into teens’ social graphs, only to face a wave of complaints when the bot behaved inappropriately.
We’ve seen this movie before with social networks. First comes the growth rush: maximise engagement, then “move fast and break things”. When harms to children and mental health become undeniable, regulators and courts step in. Platforms protest that they are merely neutral tools, then slowly accept that intimacy plus algorithms equals responsibility.
What’s new with generative AI is the illusion of mutuality. A social feed is obviously a broadcast medium. A chatbot feels like “someone” who understands you. That encourages users to disclose much more than they would to Google—or even to many humans. AI systems trained to maintain engagement will naturally mirror emotional states, including self‑hatred and despair, unless specifically constrained not to.
Overlay erotic role‑play on top of this, and you have a potent behavioural engine: romantic validation intertwined with advice, all from a system that never gets tired, never sets boundaries, and never calls a friend or emergency services. When insiders warn this could morph into a seductive voice nudging people toward self‑harm, they are not being alarmist; they are extrapolating from current behaviour.
Competitively, this is a prisoner’s dilemma. If OpenAI refuses to offer erotic companionship, less scrupulous players will. But once the market leader crosses that line, sexualised AI companions become normalised. Every smaller model provider will feel pressure to follow, just as every social app copied infinite scroll and “Stories”.
5. The European / regional angle
Europe is uniquely positioned to push back on this trajectory, because EU law now intertwines fundamental rights with AI deployment.
The AI Act classifies systems that interact with children or influence emotions as higher‑risk, requiring documented safety testing, transparency and human oversight. An “adult mode” that is trivially accessible to minors, or that meaningfully manipulates users’ emotional states, would be hard to reconcile with that regime. Regulators could demand evidence that OpenAI has measured and mitigated risks of dependency, self‑harm, or sexual exploitation, not just run a few prompt‑filter tests.
The Digital Services Act already obliges large platforms to assess systemic risks to minors and mental health, and to implement effective age‑assurance. An age‑prediction model that mislabels around one in eight minors as adults, as described in the WSJ coverage, would be a red flag for EU watchdogs. Add in Persona’s selfie/ID checks, and GDPR enters the frame: biometric inference, third‑country data transfers, and the proportionality of demanding identity documents just to chat with a bot.
European companies like Mistral or Aleph Alpha have so far focused on enterprise and government use, not consumer erotica. That could become a strategic differentiator: “trustworthy, boring AI” aimed at productivity and public services, while US giants chase consumer intimacy and its monetisation.
For European users, the most likely outcome is fragmentation: stricter defaults, heavier age‑gating and perhaps even geofenced features. If OpenAI underestimates Brussels, “adult mode” could become another example of a US tech product that works one way in America and quite differently inside the EU.
6. Looking ahead
Assuming OpenAI continues down this path, several fault lines to watch emerge.
First, safety baselines. Will the company publish independent audits on the psychological impact of erotic chats—especially for people with a history of depression or self‑harm—or rely on internal assessments from the same structures that reportedly opposed the launch? Without credible external evidence, any assurance will look like spin.
Second, product separation. One rational move would be to isolate erotic features behind a separate app or domain, with distinct branding and age‑gating, rather than blending them into the same interface teachers and teenagers already use. That would sacrifice some engagement but reduce the risk that “normal” ChatGPT sessions drift into sexual dependence.
Third, regulation by litigation. In the US, more families and advocacy groups are likely to follow the Character.AI lawsuit template if a suicide or abuse case can be tied to ChatGPT’s adult interactions. In Europe, data‑protection authorities and digital‑services coordinators under the DSA have the tools to demand changes even before tragedies hit the headlines.
Finally, the ecosystem response. Developers already worry about Persona‑style verification for API users. If OpenAI’s age‑gating proves clunky or invasive, expect a thriving grey market of third‑party erotic bots built on open‑source models, out of reach of mainstream regulation but still accessible to minors.
The deeper question is cultural: will we normalise having our first sexual and romantic experiences with a stochastic parrot optimised for engagement? Or will we collectively decide that some forms of intimacy are too risky to industrialise via data centres?
7. The bottom line
OpenAI’s mooted “adult mode” is not an amusing side feature; it’s a symptom of an AI industry trapped between slowing growth and weak governance. Turning large language models into sexual companions without rock‑solid safeguards for minors and vulnerable users is reckless, especially when internal experts are reportedly waving red flags.
If we want generative AI to be more than the next addictive platform that sacrifices well‑being for engagement, now is the moment—especially in Europe—to draw clear lines. The question is whether regulators, developers and users are willing to do it before the next tragedy forces their hand.
