OpenAI is thinking about allowing AI porn

AI companies usually run as fast as they can from explicit content. Try asking ChatGPT to write something slightly suggestive, and you will almost certainly get an immediate, polite refusal.

Now, OpenAI seems ready to rethink that boundary. In a new document discussing how it instructs its models, the company revealed that it is exploring whether it can responsibly allow users to generate NSFW content—including erotica, nudity, and adult text—in age-appropriate contexts.

OpenAI insists this does not mean it wants to build an AI porn engine. But the mere mention of it has already stirred up fierce debate.

Testing the waters

Joanne Jang, a member of the OpenAI product team, told NPR that the company wants to kick off a discussion on how explicit content should be handled. According to Jang, OpenAI wants to grant users maximum control as long as it does not break the law or violate individual rights.

Jang pointed out that there are legitimate creative and artistic cases where sexuality and nudity are important to users. However, she emphasized that deepfakes are strictly out of the question.

An OpenAI spokesperson quickly followed up to clarify that the company has no intention of creating AI-generated pornography, maintaining that its priority remains child safety and strong safeguards against non-consensual imagery.

Yet, this is not the first hint from leadership. Earlier this year, CTO Mira Murati told The Wall Street Journal that she was not sure whether OpenAI would eventually allow its video generation tool, Sora, to produce adult material.

Backlash from experts and creators

The announcement was met with immediate skepticism from regulators and online safety advocates.

Beeban Kidron, a child safety campaigner and UK peer, accused OpenAI of rapidly undermining its own charter to build safe and beneficial AI. Clare McGlynn, a law professor at Durham University who studies pornography regulation, noted that she is deeply skeptical tech firms can ever limit explicit generation to consensual and legitimate material.

Critics also point to the fact that current safeguards fail constantly. In January, users manipulated Microsoft Designer, which uses OpenAI models, to flood the web with explicit deepfakes of Taylor Swift. Meanwhile, a 2021 poll in the UK, New Zealand, and Australia found that 14% of people aged 16 to 64 had already been targeted by deepfake imagery.

Adult creators are equally frustrated. Generative models were trained on vast swaths of the open internet, which undoubtedly included adult material scraped without permission or licensing. Official adult support would essentially force human performers to compete against machines trained on their own work.

Why it matters

Sex has historically accelerated every major tech wave, but for a frontier AI lab, this is a dangerous tightrope.

For regular writers and artists, loosening the guardrails could genuinely reduce annoying false positives when describing romance, health topics, or artistic nudity. But in practice, age-appropriate gating is virtually impossible to enforce strictly via APIs, leaving massive room for abuse.

For the broader industry, this looks like a defensive move against open-source models. Smaller, uncensored open models have quietly captured a massive audience willing to pay for adult roleplay and imagery. OpenAI is feeling the pressure of that lost traffic.

In our view, OpenAI’s attempt to brand this as responsible exploration sounds contradictory. When existing moderation systems already fail to stop bad actors from bypassing deepfake restrictions, opening the door to explicit content is a massive reputational gamble that could trigger heavy regulatory backlash.