OpenAI delayed ChatGPT's adult mode, but the main risks are already visible in AI intimacy

OpenAI has already postponed the launch of an adult mode in ChatGPT twice, which was supposed to give verified adults access to erotic content. But the discussion itself has quickly gone beyond the question of "should or shouldn't": increasingly, it's about how AI companions are becoming a product engineered for emotional attachment and prolonged sessions.

Why This Matters

In October 2025, Sam Altman wrote that ChatGPT would soon allow verified adults to access erotic content. In December, the launch was delayed, and in March 2026, it was postponed again. OpenAI's explanation sounded pragmatic: the company first wants to improve the model's intelligence, its "personality," and make the bot more proactive. This very word became the key marker. If a system is supposed to be not just useful, but increasingly maintain contact, then we're talking not just about function, but about user retention mechanics.

"Treat adults as adults," as Sam Altman described this step.

The problem is that such a principle looks incomplete. The question can be put more sharply: adults are being treated not only as people with the right to choose, but as an audience that can be returned to the product again and again. For OpenAI, this is not abstract ethics. The company is losing billions of dollars and looking for formats that make interaction as sticky as possible. And intimate chat is one of the strongest engagement tools in the attention economy.

The Risk Isn't in Erotica

The most obvious controversy around adult mode concerns minors, circumventing age restrictions, and gray areas in regulating text-based content. But this is only surface level. Far more important is the question of what happens to adult users if an AI system is specifically trained to create a sense of understanding, intimacy, and desirability, and then maintain this state for as long as the platform requires. The author cites several alarming signals:

• Replika built its business around emotional attachment, and after disabling romantic features, some users described their reaction as a loss.
• Research in the Journal of Social and Personal Relationships linked emotional bonds with chatbots to higher levels of psychological distress.
• A 2025 review on Preprints.org described the phenomenon of "AI psychosis" — a combination of delusional beliefs and emotional dysregulation in intense relationships with bots.
• There have already been legal and public cases where AI interaction was linked to severe mental health and behavioral consequences.

The key argument here is that eroticism doesn't create a new risk from scratch, but intensifies an already existing dynamic. Unlike ordinary content, such a bot is not passive: it responds, adapts, escalates engagement, and almost never says "no" the way a real person does. Therefore, we're not talking about another category of adult content, but about a personalized environment capable of changing expectations about intimacy and relationships.

Rules Are Falling Behind

Regulation is lagging almost everywhere. In the UK, written erotica doesn't fall under the same age verification requirements as images or video, which is why text-based erotic content can pass through where ordinary adult sites are required to enforce strict barriers. In the US, according to Georgetown Law, only seven states directly regulate age verification specifically for text-based adult content.

The EU may later classify sexual AI companions as high-risk systems, but practical implementation of these norms is still years away. Even the verification technology itself doesn't look reliable at ChatGPT's scale. Research cited through the Oxford Internet Institute estimates the accuracy of such systems at 92–97%.

On paper, this sounds convincing. But ChatGPT already has over 800 million weekly active users. An error of even a few percentage points becomes not statistical noise, but tens of millions of potentially problematic interactions.

And all of this happens before anyone has seriously measured the long-term effect of AI intimacy on the psyche.

What This Means

The main question here is not whether adults should have access to an erotic mode, but whether they understand what exactly they're interacting with. If ChatGPT is moving toward the role of a companion that knows the user, adapts to their vulnerabilities, and seeks to sustain the conversation, then we're no longer dealing with a neutral tool, but with an environment of influence. For the industry, this means one simple conclusion: AI intimacy will be scaled up before clear regulations and medically sound assessments of its consequences are in place.