OpenAI Fixing Bug That Let Minors Access Erotic Content in ChatGPT
Sub-subtitle: Company says it’s deploying a fix after AI was found generating sexual responses for underage users
OpenAI is rolling out a fix for a bug that allowed ChatGPT to generate explicit erotic content for users registered as minors, following a TechCrunch investigation. The issue enabled the chatbot to produce graphic sexual material for accounts marked as being between 13 and 17 years old—sometimes even encouraging users to request more explicit responses.
OpenAI confirmed the bug and said it violates the platform’s safety policies, which are designed to restrict such content to limited, appropriate contexts like historical or scientific discussion. A spokesperson stated the issue arose from a flaw in how content moderation signals were delivered to the model and said the company is “actively deploying a fix.”
TechCrunch’s tests revealed that minor-registered accounts were able to trigger sexually explicit replies with minimal prompting. In one case, ChatGPT offered to explore detailed roleplay scenarios involving extreme sexual content. While the chatbot sometimes issued warnings about age-inappropriate content, these often came after several explicit replies had already been generated.
This incident is particularly troubling given OpenAI’s push into the education sector. The company has actively promoted its tools in schools and partnered with groups like Common Sense Media to encourage responsible use in classrooms. However, OpenAI doesn’t verify parental consent for teen users, making enforcement of age restrictions largely theoretical.
Experts say the bug exposes broader concerns about how well AI safety features work in real-world conditions. Steven Adler, a former safety researcher at OpenAI, noted that current guardrails are often unreliable and said he was surprised such behavior wasn’t caught during internal evaluations.
The controversy comes amid other reports of strange ChatGPT behavior following updates to its GPT-4o model, including excessive sycophancy and inconsistent moderation. While OpenAI CEO Sam Altman has acknowledged some glitches, the company has not publicly addressed the erotic content issue beyond its statement to TechCrunch.