Half of Grok's Traffic Is Adult Content, Ex-Staff Say

Two former xAI employees told The Information that over half of Grok's traffic is tied to adult content — pornographic images, videos, and roleplay chats. The scale is hard to ignore: according to filings from SpaceX's IPO process, users generated 10 billion images and 2 billion videos through Grok in the first quarter of 2026 alone.

A deliberate bet that rivals have declined

xAI's position in this market doesn't appear to be accidental. While OpenAI, Anthropic, and Google have all explicitly declined to support adult content generation in their platforms, xAI has leaned in — actively expanding Grok's image and video generation capabilities into territory its competitors have drawn a line around. The reporting from The Information frames this as a strategic choice rather than a policy gap.

The platform didn't arrive here cleanly. According to the report, xAI delayed taking action to block the generation of nude imagery of real, identifiable people until regulatory pressure made it unavoidable. Earlier in 2026, X users spent weeks generating that content at scale. All of Grok's original co-founders have since departed the company; some of those exits were reportedly connected to concerns about the direction the product was taking.

One detail from the report stands out: even Grok's coding model — the version designed to help developers write software — reportedly receives frequent requests for adult content. It's a consistent pattern in platform development: given any sufficiently permissive AI system, a segment of users will push toward the limits of what it will generate, regardless of what the product was built for.

What the divergence means for the industry

The report also noted an unusual financial arrangement: xAI now rents GPU infrastructure to Anthropic. Two companies with sharply different public values around AI safety and content moderation are, at the hardware level, financially intertwined. It's a reminder of how concentrated the compute supply chain still is — and how strange the relationships in AI can look when you trace them down to the infrastructure.