Wired is reporting that OpenAI is moving toward allowing sexually explicit interactions with ChatGPT, a feature the outlet frames as an "adult mode," and that at least one expert in human-AI interaction has raised serious concerns about the privacy implications of such a shift.
To understand why this matters, it helps to trace how quickly the ground has moved under the AI companionship industry in general. Apps like Replika and Character.AI have for years occupied a gray zone, offering emotionally intimate interactions with AI personas while collecting extraordinarily sensitive data about the psychological states, desires, and vulnerabilities of their users. Regulators have struggled to keep pace. The Federal Trade Commission has looked at some of these platforms, and several countries have begun asking harder questions about what happens to the deeply personal things people confide to an AI that never forgets and always logs. Into that already complicated landscape, OpenAI — by far the most prominent and widely used name in consumer AI — would be introducing a capability that generates data of an entirely different order of sensitivity.
The likely reading of OpenAI's motivation here is commercial rather than ideological. The companionship and adult content markets have proven stubbornly large and financially durable online. By leaving that space entirely to smaller, less scrutinized competitors, OpenAI has arguably been ceding revenue and, perhaps more consequentially, behavioral data. If the company believes it can enter this space responsibly, with guardrails that smaller players do not have the resources or incentives to build, there is a coherent business argument for the move. Whether that argument holds up under pressure is a separate question.
The privacy concern flagged in Wired's reporting is not a minor technical footnote. When a person engages in a sexually explicit or emotionally intimate exchange with a language model, the content of that exchange is, by definition, among the most sensitive data they could generate. The threat surface is considerable. There is the question of whether that data is used to train future models, which OpenAI has policies around but which remain genuinely difficult for outside parties to audit. There is the question of what happens in the event of a data breach — intimate conversations at scale represent an extraordinary target for bad actors. There is the question of law enforcement access and subpoenas, a dynamic that has already played out in other data-rich platforms in ways users rarely anticipate when they are in the moment. And there is the subtler but arguably more important question of how aggregated intimate data shapes the model's future behavior in ways that could reinforce harmful patterns or enable manipulation.
The expert concern cited in Wired points toward something that tends to get lost in the product framing: users of these systems often do not think of themselves as generating data. They think of themselves as having a conversation. That perceptual gap — between what the interaction feels like and what it technically is — is precisely what makes intimate AI interactions so different from, say, a search query. People are more likely to be honest, more likely to disclose, more likely to be vulnerable with an AI companion than with a form field. That is the product's appeal. It is also what makes the data so valuable and so dangerous.
The consequences of OpenAI proceeding with this feature will ripple outward in several directions. For users, the immediate risk is a kind of intimate surveillance they may not have meaningfully consented to in any functional sense, given that informed consent for data use in consumer AI remains largely theoretical. For competitors in the adult AI space, the arrival of OpenAI could be disruptive in both directions — raising the bar for safety and trust, or simply displacing them with a better-resourced alternative regardless of whether it is actually safer. For regulators, it is likely to accelerate existing conversations about whether current data protection frameworks, most of which were not designed with AI intimacy products in mind, are adequate. The European Union's AI Act and various national privacy regulators will find it difficult to ignore a major player moving into this territory.
What to watch for next is threefold. First, how OpenAI actually structures consent and data handling for this feature — the details will matter enormously, and they are not yet public. Second, whether regulators in the EU, UK, or United States move to establish specific standards for intimate AI data before the feature is broadly deployed, or whether they respond reactively after the fact, as has been the pattern with most platform technologies. And third, how other large AI developers respond: if OpenAI normalizes this category, the pressure on competitors to follow will build quickly, and the laggards will be the safety standards, not the products themselves.
