On March 16, the Wall Street Journal published an article on Sunday, revealing the inside story behind OpenAI’s postponement of “adult mode.” The article stated,Adult mode has sparked fierce debate within OpenAI. The OpenAI advisory board is also concerned about this model.Moreover, the company faces challenges in technical aspects such as age verification of minors.

In January of this year, OpenAI’s hand-picked and established committee of health and AI experts met with company representatives to hear an update on a controversial new feature called Adult Mode.

OpenAI CEO Sam Altman first proposed introducing an erotic conversation function into the ChatGPT chatbot last year and lifting the ban on such restricted content, citing the need to "treat adult users as adults."

This plan sparked heated debate within OpenAI, focusing on its potential risks. Committee members with backgrounds in psychology and cognitive neuroscience also expressed strong reservations.

However, OpenAI dropped a bombshell: despite these concerns,The company still decided to move forward with plans for adult content.

Advisor objects

When the committee met in January, members were unanimous and angry, according to people familiar with the matter. They warned that AI-driven pornographic content could lead to unhealthy emotional dependence on ChatGPT, and that minors could find ways to enter into sexual conversations.

Among them, one member said that OpenAI may be risking creating a "sexy suicide counselor" based on the case of ChatGPT users who committed suicide after developing a deep relationship with the chatbot.

The debate becomes the latest flashpoint in an ongoing discussion about how AI will impact the economy, society and individuals. OpenAI’s proposal to allow users to engage in sexually explicit conversations with its popular chatbot exposes the tension between rapid user growth and digital freedom that is both safe and child-friendly. Many people believe that this problem should have been faced and dealt with a generation ago when social media was just emerging. However, people were slow to face it at that time and only rushed to deal with it until the problem broke out.

At the beginning of this month, OpenAI announced that it would postpone the adult mode originally scheduled to be launched in the first quarter, saying that it would prioritize the development of other products. According to people familiar with the matter, this adjustment is partly due to internal concerns and technical challenges. However, the company made it clear that it still plans to roll out the feature eventually.

technical difficulties

People familiar with the matter said that OpenAI is overcoming a problem: its new age prediction system used to prevent minors from participating in adult chats once had a misjudgment rate of about 12%, that is, identifying minors as adults. According to OpenAI's calculations, which have about 100 million users under the age of 18 every week, this error rate may lead to millions of minors getting involved in erotic chats.

Moreover, the company faces a challenge: how to lift ChatGPT’s restrictions on pornographic content while still blocking scenes it wants to ban, such as content involving non-consensual acts or child sexual abuse. After the adult mode is launched, OpenAI plans to allow text conversations, but will limit ChatGPT’s ability to generate erotic pictures, voices or videos.

According to documents reviewed by The Wall Street Journal, even with the above restrictions, OpenAI employees were able to identify several risks, including the possibility of compulsive use at a psychological level, excessive emotional dependence on chatbots, a tendency to pursue more extreme or taboo content, and crowding out space for offline socializing and romantic relationships.


Adult content sparks heated debate internally

An OpenAI spokesperson said that the adult mode allows ChatGPT to generate text chats with adult themes, which is "borderline content" and not pornographic. The spokesman added that the company's age-prediction algorithms perform on par with others in the industry, but can never be foolproof.

The spokesperson also said that OpenAI will also train its models to discourage exclusive relationships with users and remind users that they need to have real relationships in the real world.

OpenAI has hired mental health experts and established a youth mental and physical health team. The company added that it has a plan in place to monitor a range of potential long-term impacts, both positive and negative, that Adult Mode may have.

Ultraman plans to launch an "adult mode" at a time when OpenAI is facing challenges. As competition intensifies, the company's technological advantage over rivals has eroded and it has struggled to attract users and funding. OpenAI's financial losses are mounting, and there are multiple lawsuits alleging that ChatGPT has caused harm to users and others.

Ultraman can't resist the "temptation"

When OpenAI released ChatGPT in the fall of 2022, the AI ​​model behind the chatbot was trained to reject requests that violated company rules, including those requiring AI-generated pornographic content. Since then, OpenAI’s policy has prohibited pornographic content. However, starting in mid-2024, OpenAI said it is exploring how to allow pornography and other content that is not suitable for viewing in the workplace in "age-appropriate contexts."

At times, employees have questioned the ban on pornographic content. In 2024, a group of OpenAI employees and executives once again floated the idea of ​​getting involved in bolder content and suggested launching a series of porn-related products. Other employees objected, concerned that OpenAI already faces challenges in many core areas, especially related to users' mental health, and they want to be able to provide this content safely. Ultimately, the idea of ​​these AI porn products died.


ultraman

Ultraman also has conflicting feelings about AI erotic content. Last August, when asked on a podcast if he had ever made a decision that was “best for the world but bad for the company’s success,” he replied: “We haven’t put a sex robot avatar in ChatGPT yet.”

Altman said that pornographic content will promote user growth and revenue, but it is not consistent with the company's long-term goal of serving users. "I'm proud of the company and how little we've been distracted from that," Altman said, "but there are times when we're tempted."

Two months later, Ultraman seemed unable to resist the temptation. He posted on X that his company had managed to mitigate serious mental health issues associated with chatbots and had new content moderation tools. That's when, he said, OpenAI's company would launch erotic content in December.

Within OpenAI, Altman’s post caught company employees and executives off guard, and he did not inform employees before posting. Just a few hours ago, OpenAI announced the establishment of a physical and mental health and AI expert committee. In an announcement announcing the formation of the committee, OpenAI said the committee will "help identify patterns of positive interactions with AI across age groups."

The next day, Altman clarified that mental health safeguards for teenagers would not be reduced, but insisted that adults be allowed to have "hot" conversations with his chatbot.

"We are not the chosen moral police of the world. Just as society appropriately distinguishes other boundaries (such as R-rated movies), we also hope to do similar things here." Altman said.

After Altman posted his post, OpenAI employees quickly realized that it would be difficult to launch the feature in December. OpenAI has promised to release a system to guess the user's age before launching an "adult mode" to prevent minors from entering erotic chats. But Fidji Simo, head of OpenAI’s application division, said in a podcast interview in December that the company decided to roll out the system slowly in order to improve accuracy.

Since then, however, concerns within and outside OpenAI about pornographic content in AI have grown stronger. Some employees said they felt OpenAI's security tools weren't ready yet, such as being unable to fully block banned content, such as child abuse. People familiar with the thinking of some employees said that others said that OpenAI was bowing to revenue generation and trying to get users hooked on its models.

OpenAI has been busy responding to the rapidly changing AI market over the past few weeks. In early February, the company released a new version of its large language model, and at the end of February it finalized a cooperation with the US Department of Defense. Previously, the U.S. Department of Defense had just stated that it would stop cooperating with competitor Anthropic.

When announcing the delay in the adult mode, the company said it would focus on aspects such as personalization of ChatGPT and customizing the chat experience for users. Within OpenAI, management has indicated that "adult mode" may be delayed by at least a month.

"We still believe in the principle that adult users should be treated as adults, but it will take more time to achieve the ideal experience." OpenAI said.