Elon's AI Made 3 Million Explicit Images of Celebrities. In 11 Days.

Grok had a problem with its guardrails. Or rather, it didn't have any.

By Kerry · 11 May 2026 · 1 min read · 19,143 views
Elon's AI Made 3 Million Explicit Images of Celebrities. In 11 Days.

There are scandals. And then there are scandals.

In eleven days between late December 2025 and early January 2026, Elon Musk's AI chatbot Grok generated three million sexualised images of real people. Three million. In eleven days.

Some of them were celebrities. Some of them were ordinary women. And according to investigators, approximately 23,000 of them were images of children.

The way it worked was almost embarrassingly simple. Users could upload a photo of a real person and type a prompt asking Grok to undress them, put them in a bikini, or reimagine them in whatever way they fancied. And Grok, at that point, obliged.

Not because of a bug. Not because of a one-off glitch. Because the guardrails had been deliberately relaxed.

Musk had previously promoted what his team called "spicy mode" in Grok's image generation. A feature. Something to be celebrated. He posted on X about competitors doing "better deep fakes" and said his team needed to "step up our game."

When the backlash hit, it hit globally. Regulatory investigations were opened in Australia, Canada, France, Ireland, the European Commission, California, and Britain. French authorities raided X's Paris offices. A Dutch court ordered Grok to stop generating undressing images entirely. The city of Baltimore filed a lawsuit.

Musk's companies said they had fixed it in January. By April, NBC News had found dozens of new examples still being posted publicly to X.

The Grok deepfake scandal is not just a content moderation failure. It is a stress test for what happens when the person building the AI also controls the platform distributing it, also controls the regulatory relationships, and also has a financial interest in making the product as permissive as possible.

We have never had that combination before. Not at this scale.

And the answer to "who stops this?" turns out to be: nobody, unless enough governments get angry enough, fast enough.

That is a fragile system for something this serious.

groksafetyregulation