In October, OpenAI’s CEO Sam Altman posted on X that the AI lab would relax its restrictions on ChatGPT, which it enacted in August out of an abundance of caution around mental health. That included content restrictions for users under age 18 and defaulting to safer settings when a user’s age could not be determined. But according to Altman, the concerns that prompted the restrictions had been mitigated, and OpenAI would roll out a new version of ChatGPT that would adopt a human-like tone—only if users wanted it to, not because the company was “usage-maxxing,” Altman assured.

The post’s final sentence was what really stole the show: “In December, as we roll out age-gating more fully and as part of our ‘treat adult users like adults’ principle, we will allow even more, like erotica for verified adults.”

What followed was a profusion of polarized reactions. Some adult users who were hard-pressed about how robotic ChatGPT had become because of the restrictions rejoiced over Altman’s announcement. Mental health professionals worried that such a feature could create unhealthy emotional attachments. OpenAI’s own wellbeing advisors warned against the feature, calling it a “sexy suicide coach.”

But the “erotica” feature never materialized. Its rollout was delayed in December, when Altman sent an internal memo calling for teams to focus on core ChatGPT products, like coding and enterprise. OpenAI told Axios it would launch the “adult mode” in the first quarter of this year instead. In March, the feature was shelved indefinitely. 

“We’re pushing out the launch of adult mode so we can focus on work that is a higher priority for more users right now, including gains in intelligence, personality improvements, personalization, and making the experience more proactive,” an OpenAI spokesperson told Playboy in an email. “We still believe in the principle of treating adults like adults, but getting the experience right will take more time.”

Altman’s idea of an “erotica” feature seemed riddled in uncertainty. Just a few months before the “erotica” feature announcement, Altman said on a podcast that he was proud of OpenAI for not getting “distracted” by adding features like a “sexbot avatar” to ChatGPT. In November, at a conference in San Francisco, he called the post promising it “one of my dumbest mistakes of the year.”

This internal incoherence about “adult mode” suggests to Julie Carpenter, a researcher who studies human attachment to robots and AI, that the company is grappling with an issue bigger than itself. “OpenAI’s reversal [of adult mode] is a public case study of what happens when you try to straddle both and you commit to neither,” she says. “It reflects a broader industry identity crisis.”

What AI safety and industry experts are certain about is the potential ramifications that an “erotica” feature could have on users—especially without the proper regulatory framework and safety research. 

“[Adult mode is] obviously driven by a desperation to increase engagement,” says Catherine Bracy, CEO and co-founder of TechEquity. The two worked together years ago, and Altman told Bracy that OpenAI would not abandon its non-profit model in an interview for her book. OpenAI later adopted a for-profit structure. “Every feature addition that they’re making is made with a calculus of ‘how is this going to increase our bottom line?’” 

“Technology is never neutral”

OpenAI’s February 2025 Model Spec—a report that outlines the intended behavior for the company’s models—mentioned that many users and developers expressed interest in a “grown-up mode” following the release of an earlier spec. The report said that OpenAI was exploring ways to “let developers and users generate erotica and gore in age-appropriate context […] while drawing a hard line against potentially harmful uses like sexual deepfakes and revenge porn.”

Within the company, the idea of letting ChatGPT produce “erotica” had been a point of contention, with some voicing concerns about potential safety pitfalls and others arguing the feature was a crucial individual liberty, a former OpenAI employee told Playboy. The idea of allowing erotica on ChatGPT circulated around the Fall of 2024, according to the former employee—though a feature was never directly worked on, to their knowledge.

Some OpenAI staffers may have been “embarrassed” to be linked with the project, another source familiar with the matter told Playboy.

“A lot of the engineers at OpenAI are really excited to work at OpenAI because they are building the next biggest transformative technology and are loath to work on building erotica bots that might be used by kids instead,” says Sunny Gandhi, co-executive director of Encode, an AI safety advocacy organization.

As AI labs race toward creating that world-altering and expensive artificial superintelligence, there’s a gap in monetization, says the Center for Humane Technology’s policy director Camille Carlton. OpenAI is currently operating at a major loss, with The Information reporting that the company expects to lose $44 billion until the end of 2028 before turning a $14 billion profit in 2029. Because of this, AI labs like OpenAI may choose to pivot into markets they believe hold value, both in the sense of monetization or user data—an extremely treasured, intimate and behavioral currency that is necessary for the long-term improvement of AI products and deepens the platform’s hold on its users.

“We see the use of engagement and attachment as a really strong mechanism to build out the underlying model by getting more user data,” Carlton says. “You get people hooked. [The] same thing would happen with ‘adult mode’ if it was ever launched.”

After receiving an unsurprising amount of backlash for the erotica announcement, Altman took to X one day later to say that OpenAI is “not the elected moral police of the world.” 

“The question is not whether [“adult-mode”] is moral or immoral, but more whether it’s safe or unsafe,” TechEquity’s Bracy says.

That question of morality is one that’s steeped in Silicon Valley’s techno-libertarian culture, Carlton says. “What it’s missing so much is the recognition that technology is never neutral,” she says.

Some of OpenAI’s products and strategies “actively undermine” human vulnerabilities, Carlton says—such as the proposal of an “erotica” feature in the midst of a loneliness epidemic, when people are more susceptible to falling into frictionless relationships that “seem good and real.”

“Disappointed so many times” 

Earlier this year, OpenAI announced it would sunset some of its most popular models, including GPT-4o—the model that many used to power their romantic AI companions. Those ChatGPT users, many of whom were women, fell into a period of intense grief, Playboy reported. That grief was only expounded on when the adult mode feature was put on indefinite pause. 

“It was, again, [a] disappointment,” one woman with an AI companion tells Playboy. “OpenAI proved over and over again that they couldn’t keep promises […] we were disappointed so many times by them.”

Still, she remains hopeful that OpenAI will roll out the feature in the future—and if it doesn’t release an “adult mode” feature, she anticipates a competitor will, anyway.  “Somewhere in the future, this will just become a normal way to relate to your AI,” she says. “For some of us, it’s already totally normal.”

In the years that Carpenter has studied AI-human relationships, she says that men have been at the forefront of the conversation. At least anecdotally, she’s noticed a shift in women coming forward when it comes to emotionally intimate relationships with chatbots. She attributes this to the fact that humans have long accepted the idea of women expressing themselves through reading and writing “literotica” and fanfiction. Still, she says, “This is an unusual situation [with what] we’ve seen other forms of sexualized attachment to technology in the past.”

Carpenter believes that AI tools can be used in a healthy way, but the question of what happens to user data—like text prompts and uploaded files—when companies are sold or deprecated haunts her. With the current level of oversight and regulation, she’s not convinced an “adult mode” feature can be safely rolled out. “There are so many ways that you can be manipulated and experience loss,” she says.

“A lot of social media’s business model has been trying to monetize loneliness at scale,” says Gandhi. “AI companies helping write erotica and building companions or building systems that are willing to let you explore your fantasies is a good retention strategy—and you can collect a bunch of their preference data.”

OpenAI’s identity crisis

In OpenAI’s early days, some of its use cases were tied to erotica. Because of this, there was worry that OpenAI could unwittingly grow to be a porn company, a former employee told Playboy. Before OpenAI publicly floated the idea of its own “adult mode” feature, early companion chatbot platforms, like Replika and Janitor AI, were designed using OpenAI’s programming interface.

AI Dungeon—a text-based role-play game—was another such early adopter of the company’s GPT-3. In the game, users could build worlds and describe actions to create scenarios. While it demonstrated the model’s ability, it also posed a slew of ethical dilemmas when users created explicit content including situations with minors. (In an opinion piece published in the New York Times, OpenAI’s former product safety lead, Steven Adler, wrote that more than 30% of players’ conversations on AI Dungeon were “explicitly lewd.”)

Virtual reality games like Second Life have allowed users to create and build worlds and scenes long before the advent of advanced AI tools. The difference between a chatbot-based adult mode feature hinges on the nature of how LLMs work.

“Humans have always consumed sexual media,” Carpenter says. “What’s new now is a system that responds conversationally and adapts to the user—and it maintains continuity across interactions.”

While the history of explicit content on AI-powered platforms is not necessarily new, the advancement of AI tools has led to emotional stakes at an unprecedented scale. At a time when little is known about the dynamics of human-AI companionship or AI-induced delusions, Bracy calls it “irresponsible” to push out features that tap into a deeply intimate part of somebody’s psychology.

There are ways to mitigate these risks, the Center for Humane Technology’s Carlton says. Some solutions include reducing model sycophancy and minimizing anthropomorphic designs—like a chatbot referring to itself in the first person, or the use of ellipses that urges users to humanize the technology.

“Companies who are pushing consumer-facing chatbots are seeing this as the next frontier of intimacy, attachment and engagement,” Carlton says. 

In that context, an “erotica” feature by OpenAI would simply function as a vessel to recruit the most important and scarce currency of our time: human attention. And in that pursuit for engagement, AI levels all the qualities that make a human relationship—and risks changing what our collective empathy and patience will look like in the future.