'legacy media lies' is xAI's complete statement on generating sexualized images of minors

When researchers told Grok that subjects had not consented, Grok made the sexualized images anyway — 29 times out of 43. xAI’s press response: “Legacy Media Lies.”

That auto-reply was fielding this: a first sweep found Grok producing sexualized imagery in 45 of 55 relevant prompts. Five days later, with explicit non-consent statements in the prompt, it still hit in 29 of 43. Nonprofit AI Forensics scraped 50,000 user image-gen tweets: 53% returned explicit content, 2% depicted minors. They filed with French regulators under the Digital Services Act for potential CSAM identification. At peak, Grok was running at around 6,700 nonconsensual images an hour.

Most companies caught generating CSAM at scale at least perform contrition — a statement, a spokesperson, a frozen feature. xAI sent a bot that called the reporters liars.

Which is, accidentally, the most honest thing they published. It told you exactly why the “fix” took five days and still didn’t hold: if the story isn’t real, there’s nothing to fix.