ChatGPT Adult Mode: What OpenAI Has Actually Confirmed

As of September 15, 2026, no OpenAI communication we can find confirms the specific claim circulating among ChatGPT users: that a promised 'adult mode' was paused indefinitely in March 2026. The only verifiable record is OpenAI's own policy update from September 16, 2025, which described plans to treat age-verified adults differently from teens, including looser rules for mature content such as erotica, explicitly conditioned on first shipping a working age-prediction and verification system. OpenAI never announced a firm launch date, and it has not published any statement revising, pausing, or reviving that plan since. For the four platforms we track in the AI Sex Chat & Roleplay category, this uncertainty sits alongside a separate, measurable pattern: most of what those services disclose about censorship, memory, and exit terms is self-reported rather than independently observed.

OpenAI's own account never promised a launch date, only a precondition

OpenAI's September 16, 2025 post on teen safety, freedom, and privacy is the closest thing to an official statement on this topic. It explains that the company intended to treat verified adults differently from users it believes are minors, tying any loosening of content restrictions, including allowing more mature material, to the deployment of age-prediction and verification tools. Crucially, OpenAI framed this as a future-conditional policy, not a shipped or dated feature: the company stated that changes for adults depend on first getting age verification working reliably, not the reverse.

Searching for a primary OpenAI source, a blog post, help-center article, or official statement, that specifically confirms a 'March 2026' pause of this plan, or that provides any revised timeline, turns up nothing. That absence does not prove the pause did not happen; it means the claim, as framed by users, cannot currently be checked against an OpenAI communication. What can be said with confidence is narrower: as of the September 16, 2025 statement, the feature was conditional and undated, and no OpenAI update since then has moved that timeline forward, backward, or anywhere at all.

A year of silence leaves the stated precondition unresolved

Between OpenAI's September 16, 2025 announcement and our September 15, 2026 review date, roughly a year has passed without a follow-up statement narrowing or confirming the plan. OpenAI's own framing means two separate things would need to happen before any 'adult mode' could ship: the age-prediction and verification system would need to be built and deployed reliably, and only then would the content-policy change for verified adults follow. Neither OpenAI's original post nor any later communication we can locate states that the verification system itself has shipped, been tested, or been delayed.

This matters for how the rumor should be read. A 'pause' implies something was running and then stopped; what the public record actually shows is a conditional plan that was announced once, tied to an unfinished prerequisite, and never given a public status update. Users interpreting silence as either confirmation of a pause or as evidence of an imminent launch are both going beyond what OpenAI has said. Until OpenAI issues a new statement, the accurate description is that the feature's status is undefined, not paused or cancelled in any officially documented sense.

Across the four platforms we track, declared claims still outweigh independent tests

While ChatGPT's plans remain undocumented, the four platforms tracked in the AI Sex Chat & Roleplay category as of September 15, 2026 illustrate a related transparency gap. Our censorship axis, which measures how closely a platform's actual content filtering matches what it states in its own policies, averages 5.1 across these four platforms, with three of the four scores based on declared information and only one inferred from indirect signals, none independently observed. The persistence axis, covering how reliably a platform retains conversation memory across sessions, averages 4.3, again split between declared and inferred scores with zero drawn from direct observation.

The pattern repeats on credits, where pricing-to-usage value averages 3.5 with a single declared score and three inferred, and on exit terms, which measures how easy it is to cancel or get a refund: that axis averages 2.9, ranging from a low of 0 to a high of 5.5, built from three declared scores and one inferred. Only data retention, at an average of 3.6, is fully declared across all four platforms, meaning every score there reflects what each service states in its own policy rather than a hidden or tested figure. None of the five axes we track for this category currently include a directly observed score, which is worth stating plainly rather than glossing over.

What to check before paying for any platform claiming an 'adult mode'

Given that OpenAI's own plan remains undated and unconfirmed beyond its September 2025 framing, and that the four alternative platforms we track lean heavily on self-reported rather than tested claims, anyone paying for adult-oriented AI chat access should verify a few things directly rather than relying on marketing language or forum rumors.

• Look for a dated, named source (a company blog post, help-center page, or press statement) before assuming any 'adult mode' announcement is current or still active.
• Check whether a platform's censorship or content-filtering claims are labeled as tested by the reviewer or simply declared by the platform itself.
• Confirm what happens to stored conversations and account data after cancellation, since data-retention terms in this category currently average 3.6 and are drawn entirely from platform-declared policy text.
• Review exit and refund terms before subscribing, given this axis shows the widest spread of the five tracked, from 0 to 5.5, indicating meaningfully different cancellation experiences across otherwise similar services.
• Treat any claim of a specific launch date or 'confirmed pause' for a major provider's feature as unverified unless it traces to a primary statement from that company.