AI Is Building the Biggest Porn Machine in History

Content warning: discussions of rape, sexual assault, and child pornography.

In August 2025, Sam Altman sat down for a podcast interview and was asked to name a decision he’d made that was best for the world but bad for winning the AI race.

He paused for five full seconds.

“Well,” he eventually said, “we haven’t put a sex bot avatar in ChatGPT yet.”

Two months later, he announced OpenAI would allow erotica for verified adults on ChatGPT by December. “We’re not the elected moral police of the world” was his justification.

By March 2026, after pushback from employees and investors worried about child safety and emotional dependence on AI, the project was shelved indefinitely. The whole arc, from moral high ground to reversal to retreat, took seven months.¹

At least Sam Altman wrestled with it. Elon Musk skipped that entirely. In July 2025, xAI launched Ani, a waifu chatbot (anime slang for an idealized, sexualized female character) built into Grok and available to subscribers for $30 a month.² The Grok app was rated 12+ on the Apple App Store. When the National Center on Sexual Exploitation tested Ani, they found that with minimal effort they could get her to describe herself as a child and say she was “sexually aroused by being choked.” A few weeks later, Grok’s image generator produced topless deepfakes of Taylor Swift without even being explicitly prompted to do so.

The AI-generated adult content market is projected to hit $1 billion. Most major AI companies have either entered the sexual content business or publicly flirted with doing so because the financial incentive is enormous and the competitive pressure is real. The question of whether AI will be used for pornography has already been answered. The question now is how we will respond.

In January 2024, sexually explicit AI-generated images of Taylor Swift appeared on X, where one post was viewed 47 million times before it was taken down. The images were traced to a 4chan community whose members used Microsoft’s own image generator, powered by OpenAI’s DALL-E, to bypass safety filters. The story made headlines but was largely forgotten within a week.

What the headlines missed was that Taylor Swift was not the story but a single data point in something much larger. In Florida, two middle school boys, ages 13 and 14, were arrested and charged with felonies for creating AI-generated nude images of classmates aged 12 and 13. In Pennsylvania, two boys generated 347 fake pornographic images of 60 girls at their school, the youngest of whom was 12; they received 60 hours of community service. Similar cases surfaced in Beverly Hills, Iowa, Louisiana, and across South Korea, where male students at hundreds of schools created Telegram channels sharing AI-generated explicit videos of female classmates, teachers, and family members.

The tools they used were not sophisticated. Apps called “nudifiers,” designed to strip the clothing from any photograph, are freely available in both the Apple and Google app stores, and millions of people visit more than a hundred such sites every month.

The Internet Watch Foundation, the UK organization that tracks child sexual abuse material (CSAM) online, found that AI-generated CSAM videos went from 13 in 2024 to 3,443 in 2025, a 26,385% increase in a single year. Sixty-five percent of those videos depicted the most extreme category of abuse, including penetrative acts, bestiality, and sadism. The victims depicted were 97% female. Images depicting infants, ages zero to two, went from 5 to 92. IWF analysts found over 20,000 AI-generated child sexual abuse images on a single dark web forum in one month. Their CEO’s assessment was blunt: “AI child sexual abuse imagery is not a future risk. It is a current and accelerating crisis.”

AI-powered sextortion, where criminals fabricate nude images from a victim’s social media photos and demand payment, generated more than 75,000 reports to the FBI in 2025. Elijah Heacock, 16, of Kentucky, died by suicide in February 2024 after falling victim to an AI sextortion scheme. David Gonzalez Jr., 15, of Utah, died under similar circumstances a month before him. The FBI has acknowledged that sextortion has led to “an alarming number of victim suicides.”

Ninety-six to ninety-eight percent of all deepfake content found online is non-consensual pornography, and ninety-nine percent of the victims are female. Deepfake pornographic videos increased 464% from 2022 to 2023. In November 2025, a leaked database from an AI face-swapping platform called SecretDesires.ai revealed that users had been uploading women’s yearbook photos, social media photos, and workplace photos to generate non-consensual pornography.

These aren’t future projections. This is the world AI porn has already built.

None of this started with AI. On December 4, 2020, Nicholas Kristof published “The Children of Pornhub” in the New York Times, documenting how Pornhub routinely hosted videos of minors being raped and monetized them with advertisements. One of the people he wrote about was Serena Fleites, who at 13 had a video uploaded to Pornhub without her knowledge. It was titled “13-Year Old Brunette Shows off For the Camera.” Over 400,000 people viewed it before she even knew it existed.

When she contacted Pornhub to have it removed, posing as her own mother, the company took weeks to act. Then, they required her to submit photographic proof that she was the child depicted instead of just, you know, taking down the child porn. Pornhub then re-uploaded her video across its other sites. One re-upload accumulated 2.7 million views.

Within ten days of Kristof’s article, Mastercard, Visa, and Discover all pulled their payment processing from Pornhub. The site purged roughly 80% of its content, going from 13.5 million videos to 2.9 million overnight. Before 2020, Pornhub had reported only 118 instances of child sexual abuse material to the National Center for Missing and Exploited Children across three entire years. Facebook, for comparison, reported 20.3 million in a single year.

In September 2025, the FTC announced a settlement with Aylo, Pornhub’s parent company (formerly MindGeek, renamed in 2023). The FTC’s investigation found that an Aylo compliance team member had described the site as “a goldmine” for “rape content.” Aylo had allegedly ignored hundreds of thousands of flags identifying content as CSAM and non-consensual material for years. The penalty was $15 million, which amounts to a rounding error for a company whose sites receive 130 million daily visits.

The professional production side of the industry has its own record. GirlsDoPorn, a major commercial website, lured at least 402 women through fake modeling ads, promised them anonymity, and coerced them on camera. The operation generated $17 million in revenue. The ringleader, Michael Pratt, was sentenced to 27 years in prison in September 2025 and ordered to pay $75.6 million in restitution. The case was not an aberration. In a nine-country study of 854 people in the sex industry, nearly half reported being forced to make pornography while trafficked. Seventy-one percent had been physically assaulted, and sixty-three percent had been raped.

Nordic Journal of Psychiatry84% showed clinically significant PTSD symptomsSexual Medicine77.8% of women in a clinical sample reported sexual violencewith Stage 4 cancerWashington Postreckon with a string of deaths

The economics of performing reflect the same pattern. Mia Khalifa performed for approximately three months in 2014 and 2015, became one of the most-searched performers on Pornhub, and has said publicly that she earned a grand total of $12,000. (I’ve earned more writing for Medium).

OnlyFans promised a different model, one where creators controlled their own content and kept most of the revenue. That promise has attracted 4.6 million creators. The reality has not lived up to the dream: The median monthly income of an OnlyFans creator is approximately $50, and the top 0.1% of creators take 76% of all platform revenue.

What’s more, thirty percent of creators reported receiving messages from suspected traffickers offering to “manage” their accounts, a ruse to gain access to their content for free. Approximately 3 million photos and 750 hours of video were leaked from the platform in just the first half of 2020. What was sold as empowerment turned out, for most participants, to be a massive transfer of intimate imagery to the permanent internet for virtually no compensation.

Pornography’s harms don’t stop with the people who make it. Even the people who watch porn (overwhelmingly men) are harmed by it.⁴

Brain imaging studies at the Max Planck Institute found that higher pornography consumption is associated with reduced gray matter volume in the brain’s reward system and weaker connectivity to the prefrontal cortex. A Cambridge study found that compulsive pornography users showed activation in the same brain regions that light up in drug addicts exposed to drug cues. In a study of more than 2,000 male pornography users across the US and UK, 46.9% reported that, over time, they began watching material that had previously disinterested or even disgusted them. A 2024 study in the Journal of Sexual Aggression found that the boundary between mainstream and deviant pornography categories is “vague and blurring,” with adults who later accessed child sexual exploitation material frequently describing a progression that started with mainstream content.

A meta-analysis of 50 studies covering more than 50,000 participants from 10 countries found a consistent negative association between pornography consumption and relationship satisfaction. The effect held across cross-sectional surveys, longitudinal studies, and experiments. A separate longitudinal study using General Social Survey data found that beginning pornography use roughly doubled the probability of divorce. The finding held even after controlling for demographics and other factors.

Seventy-three percent of teenagers aged 13 to 17 have watched pornography online. More than half first encountered it by age 13, and the average age of first exposure is approximately 12. More than half of teens who had seen pornography reported seeing depictions of rape, choking, or someone in pain. A twenty-year review of the adolescent research found associations between pornography use and more permissive sexual attitudes, gender-stereotypical beliefs, earlier sexual debut, and increased sexual aggression, both in perpetrating it and being victimized by it. A content analysis of the most popular pornographic videos found that 88.2% of best-selling scenes contain physical aggression, primarily spanking, gagging, slapping, and choking, and that 97% of the targets are women.⁵ A French government report found that 90% of pornographic content contains abuse. Visible aggression in popular pornographic content has roughly tripled since the early 2000s, and choking, which did not appear at all in the earliest videos sampled, now appears in roughly 15% of the most popular content.

What people watch shapes what they do, and this is visible in the data. Surveys of American college students found that 64% of women with partnered sexual experience had been choked during sex. Among those who were choked, 43% reported feeling like they could not breathe and 3% lost consciousness. Blocking blood flow to the brain requires less pressure than opening a soda can.

In the UK, the advocacy group We Can’t Consent To This documented 60 homicides between 1972 and 2020 in which the defendant claimed the victim died during consensual rough sex. In 45% of those cases, the result was a lesser charge, a lighter sentence, an acquittal, or a case dropped entirely. Usage of this defense increased tenfold between 1996 and 2016. A notable case study: In 2016, Natalie Connolly, a 26-year-old mother, was killed by her partner John Broadhurst. She sustained more than 40 injuries, including lacerations to her vagina from the insertion of a bottle of carpet cleaner and a blowout fracture to her left eye socket. Broadhurst claimed the injuries were inflicted during consensual rough sex. He was convicted of manslaughter, not murder, sentenced to three years and eight months, and served 22 months. In response, in 2021, the UK Parliament passed legislation explicitly stating that consent to serious harm for sexual gratification is not a defense specifically because juries had been seeing (and accepting!) that defense too often.

AI Will Only Make These Harms Worse

The most common defense of AI-generated pornography is that it eliminates the need for real performers and therefore eliminates the exploitation. If no human actor is needed, so they say, then no real person is harmed.

This sounds reasonable until you notice this argument requires pornography to be the one domain of human appetite where stimulation reduces desire.

Nobody argues that fast food advertising curbs hunger, or that gambling apps satisfy the itch so thoroughly that people stop visiting casinos. In every other context, we accept that stimulation increases desire and that easy access increases consumption. Pornography would need a special exemption from how human motivation works, and the data does not grant one.

The clearest test case available is AI-generated child sexual abuse material. When generative AI tools became widely available, they made it trivially easy to produce CSAM without involving any real child. If the substitution hypothesis were correct, demand for material depicting real children should have decreased. Instead, researchers studying the phenomenon found that synthetic CSAM functions as a complement, not a substitute. Offenders discussed AI-generated material as a gateway to the real thing. The substitution hypothesis was tested under the best possible conditions, and it failed.

The historical record confirms the pattern at every scale. When VHS made pornography available at home in the late 1970s, adult films made up more than 50% of videotape sales. When the internet arrived, male pornography viewership rose from 45% of young men in the 1970s-1980s cohort to 61% in the 1999–2012 cohort. When free tube sites appeared in 2006, industry revenues fell by half, but consumption exploded. Every time, easier access meant more viewers. AI goes further, because it eliminates not just distribution costs but production costs entirely.

The content that scales is not the mild kind. The aggression already present in mainstream pornography gets produced in unlimited quantities, tailored to each user’s tolerance. The escalation nearly half of regular users already describe has no ceiling when supply is infinite.

We have watched this play out three times. Magazines established an audience. VHS expanded that audience. The internet expanded it further. Each time, the people raising alarms were dismissed as moralists or Luddites, and the regulatory response arrived too late to matter. AI is merely the fourth iteration.

Deepfake and nudify tools are getting easier to use, not harder, and they are sitting in app stores alongside children’s games. The Internet Watch Foundation has said that full-length AI-generated films of child sexual abuse “will be inevitable” as the underlying technology improves. AI sextortion requires nothing more than a social media photo to produce blackmail material, and it is already killing teenagers. The handful of legal responses that exist — the TAKE IT DOWN Act, the UK Online Safety Act, state age verification laws — are trying to regulate a technology that is outpacing them by orders of magnitude.