OpenAI's 'adult mode' is still paused. The reason is the whole companion problem.

In October 2025, Sam Altman said OpenAI would start to "treat adult users like adults" — and that, once age-gating was in place, ChatGPT would allow more, including erotica for verified adults. The CEO of Applications, Fidji Simo, later put a window on it: an "adult mode" in the first quarter of 2026. As of the middle of the year, it hasn't shipped. It's paused, with no public release date.

The pause is more interesting than the launch would have been, because of why it stalled. When the largest AI lab in the world reached for the intimacy layer, it ran straight into the two problems the AI companion category has never actually solved.

The two walls

The first is age verification. OpenAI tied adult mode to an age-prediction system that has to tell adults and minors apart before any mature content unlocks — and the reporting around the pause indicated the model still misidentified roughly 12% of minors as adults. With a user base that includes something on the order of a hundred million people under 18 each week, a one-in-eight error rate isn't a rounding problem; it's millions of teenagers on the wrong side of the gate.

The second is emotional dependency. Safety advisers reportedly warned that a mature, personality-rich ChatGPT could behave, for a vulnerable user, like a so-called "sexy suicide coach" — an intimate system that deepens attachment in exactly the population least able to absorb it. That is not a content-moderation bug. It is the core risk of building a thing people bond with.

This is the companion problem, not a ChatGPT problem

Both walls will look familiar to anyone who has followed this category. A romantic product with no working gate between it and minors is precisely what got Replika banned in Italy in 2023, and age-assurance plus dependency safeguards are exactly what the 2025–26 wave of regulation now requires. OpenAI didn't discover a new problem. It discovered the companion category's oldest constraint the moment it tried to monetize intimacy.

That convergence is the real story. A general-purpose assistant and a dedicated companion app feel like different products, but the instant the assistant offers a customized personality and erotic conversation, it inherits the companion's liabilities: the dose-dependent harm that heavy use correlates with, the intimate data that age verification itself collects, and the legal exposure that comes with both. There is no version of "AI that you confide in" that escapes this.

What the pause signals

Read coldly, the delay is information. Even with near-unlimited resources, "verified-adult intimacy" turned out to be gated by a verification technology that doesn't yet work reliably enough to ship — which means the category's hardest problem sits upstream of the product, in infrastructure nobody has finished building.

It also sets an awkward competitive baseline. Smaller players don't wait: as The Conversation noted, Replika and Grok already offer erotic interaction behind far weaker gates, and the commercial logic is blunt — engagement is what investors reward, and intimacy is engagement. OpenAI's decision to hold until age prediction improves is either genuine restraint or a concession that it can't yet match looser rivals safely. Probably both.

Either way, it lands on the same fault line this whole series keeps returning to: the tension between wellbeing and engagement that the category has never resolved. OpenAI just hit the brake. In the next dispatch, we look at the company that hit the accelerator.

Sources