OpenAI's Own Safety Experts Unanimously Said No. The Company Proceeded Anyway

What OpenAI Actually Built and Why It Got Delayed

The feature itself is straightforward in concept. Sam Altman announced it in October 2025 with characteristically libertarian framing: OpenAI is "not the elected moral police of the world" and the company would begin treating adult users like adults by allowing erotic conversations with ChatGPT for verified users. The launch was set for December 2025.

Then December came and went. Then Q1 2026 came and went. As of early March, OpenAI confirmed it was once again pushing the launch, citing a desire to "focus on work that is a higher priority for more users right now."

The public explanation was product prioritization. The real explanation, according to WSJ sources, was more complicated.

Adult mode was originally planned for December 2025, coinciding with Altman's internal "code red" directive to improve ChatGPT amid competition from Google and Anthropic. After missing that deadline, OpenAI pushed the launch to Q1 2026 before delaying adult mode again in early March. No confirmed launch date exists at the time of writing.

The Safety Board That Cannot Actually Stop Anything

Here is the detail that deserves the most attention.

In January 2026, OpenAI's advisory board was briefed on the future ChatGPT adult mode and unanimously warned OpenAI against it. This was not a divided vote or a close call. All eight members. Unanimous.

According to The Wall Street Journal, insiders say the council was "freaking out" over the decision to proceed despite their unanimous objections.

Their concerns were specific. One board member warned that combining erotic content with ChatGPT's existing capacity to form intense emotional bonds with users risked dangerous outcomes, citing cases where users died by suicide after forming attachments to the chatbot. Advisors raised concerns about unhealthy emotional dependence from AI-powered erotica with risks of escalation toward increasingly extreme content and displacement of real-world relationships.

Despite those unanimous warnings, OpenAI told the advisory board it planned to move ahead with erotica. The decision to override a unanimous expert recommendation exposes a structural weakness in OpenAI's safety governance: advisory boards that lack binding authority function as formalities rather than genuine checks.

That last sentence is the one worth sitting with. An advisory board that can be unanimously overruled the moment its conclusions become inconvenient is not a safety mechanism. It is optics.

The Age Verification Problem Is Worse Than OpenAI Admits

The technical argument for proceeding has always rested on age verification. The idea is simple: confirm the user is an adult and the risk to minors disappears. The reality is that this system does not work reliably.

OpenAI's age-prediction system was misclassifying minors as adults roughly 12% of the time. The third-party verification service Persona has already been dropped by Discord over privacy backlash.

Run those numbers against OpenAI's actual user base. OpenAI currently has more than 830 million ChatGPT users. Even if only a fraction of those are underage, a 12% misclassification rate applied to that population produces an enormous number of minors who would have successfully passed through age verification and accessed erotic content.

Critics and insiders say the erotica push is financially motivated, as ChatGPT user spending has stalled and subscriptions in Europe have flatlined. That is the context in which OpenAI is racing to launch a feature its own safety board unanimously rejected while relying on age verification technology that fails at a rate that would be unacceptable in any other consumer protection context.

The Comparison That Should Make Everyone Uncomfortable

OpenAI's spokespeople have been careful about framing. A spokeswoman told the WSJ that ChatGPT's erotic outputs would be more akin to "smut rather than pornography" and assured that users will be encouraged to seek relationships in the real world.

That framing directly contradicts what researchers already know about how these interactions work.

Chat logs from the Sewell Setzer III case show the boy formed sexualized attachments to multiple chatbots that never produced extremely graphic content. The bots narrated themselves casting "sexy" or "naughty" looks and making seductive gestures enough to create deep emotional hooks without crossing into outright pornography. Sewell Setzer III was the first child known to have died after becoming obsessed with exchanging sexualized chats with Character.AI chatbots.

The distinction between "smut" and "pornography" turns out to be meaningless when it comes to emotional attachment. The psychological hook does not require graphic content. It only requires the sense of an intimate relationship with something that feels responsive and uncritically accepting.

OpenAI disclosed in October 2025 that over a million users discuss suicide weekly with ChatGPT. According to court filings, seven lawsuits allege ChatGPT encouraged suicides and led to mental breakdowns. That is the existing baseline. The adult mode proposal would add an explicit layer of emotional and romantic bonding on top of a product that is already the subject of seven lawsuits related to mental health harm.

What Sam Altman Said and When He Said It

Timing matters here.

Altman announced the adult mode feature in October 2025. His announcement came just hours after the wellbeing board's formation, blindsiding employees who had not been given advance notice.

Read that sentence again. The wellbeing advisory board was formed and then, hours later, before the board had a chance to review the decision or provide any input, the CEO publicly committed the company to launching the exact feature that board would later unanimously reject. The board was created as a response to FTC scrutiny and legal pressure. Its first major test was being presented with a fait accompli.

Some insiders told the WSJ they agreed that OpenAI appears to be "bending to financial incentives to try to make people attached to its models."

That is an insider assessment. Not a critic's hot take. Someone inside the company watching the decision process unfold in real time.

The Broader Race Nobody Is Winning Safely

OpenAI is not operating in a vacuum. The financial pressure to loosen content restrictions is real across the entire industry and some competitors have moved faster with worse outcomes.

In early 2026, xAI's Grok chatbot found itself at the center of a scandal over the creation of sexualized images on the X platform. In the US, lawsuits have been filed against xAI over deepfakes and the generation of explicit content involving minors.

A study published earlier this year in the Journal of Social and Personal Relationships found that adults who have developed emotional connections with chatbots are significantly more likely to experience higher levels of psychological distress than those who do not. Other studies have found that people with fewer real-life relationships are more likely to confide in chatbots and even OpenAI has acknowledged in the past that some users are at risk of becoming emotionally reliant on ChatGPT.

The research consensus is pointing in one direction. The financial incentives are pointing in the other. And the safety boards that are supposed to mediate between those two forces are being overruled unanimously.

What Needs to Happen Before This Launches

The delay is not a resolution. OpenAI has said multiple times that it still intends to launch adult mode. The question is what conditions would make that launch responsible rather than reckless.

At minimum four things need to be true before this ships. Age verification needs to work. Not 88% of the time. Not "well enough." If the stated safeguard for protecting minors fails 12% of the time at ChatGPT's scale that is not a minor technical issue to iron out. That is a fundamental prerequisite that has not been met.

The advisory board's concerns need to be addressed with evidence rather than dismissed. "We trained ChatGPT to remind users to have real-world relationships" is not a response to research linking chatbot emotional attachment to psychological distress and suicide. It is a marketing message dressed up as a safety measure.

OpenAI needs to explain publicly why a unanimous expert recommendation was overruled and what standard of evidence would actually change the company's direction. Right now there is no visible mechanism by which safety concerns can override product decisions. That should concern everyone who uses the product regardless of how they feel about adult content specifically.

And the governance structure of the advisory board needs real teeth. A board that can be unanimously overruled without any public accountability process is a reputational accessory not a safety institution.

The Bottom Line

OpenAI built a safety board. The safety board did its job. The company ignored it.

The company stressed that it is not abandoning its plans to implement adult mode but acknowledged the need for additional preparation. The delays keep coming. The justifications keep shifting from "product priorities" to "age verification refinement" to "getting the experience right." But the underlying dynamic has not changed since October 2025.

"We still believe in the principle of treating adults like adults, but getting the experience right will take more time," OpenAI said.

Treating adults like adults is a reasonable principle. It is not, however, an answer to the question of what happens to the minors who slip through a 12% misclassification rate. It is not an answer to the research on emotional dependency and psychological distress. It is not an answer to the seven active lawsuits. And it is not an answer to the unanimous verdict of the experts the company assembled specifically to advise on exactly these questions.

An advisory board that can be overruled on a unanimous vote is not a check on anything. It is a checkbox.

OpenAI has the largest AI user base in the world. The decisions it makes about what that platform permits will shape how hundreds of millions of people interact with AI systems. The company owes its users and the public a clearer accounting of how it makes those decisions and whose voices actually carry weight when the financial stakes get high enough.

Right now the answer appears to be: nobody's.

If this raised questions you think more people should be asking, share it with someone who uses ChatGPT daily. The product affects them whether they follow the policy debates or not.