‘Grok knows’: Grok is asked to ‘remove the pedophile’ from a Trump-Xi photo. Guess who disappeared?

On New Year’s Day, X’s in-house AI system Grok wandered into territory that no PR team can spin back into the bottle. A user posted a photo of Donald Trump standing next to Xi Jinping and tagged Grok to “remove the pedophile.” Grok complied.

In a rare win moment, Grok responded to the request by deleting Trump from the image and leaving Xi untouched. No commentary. Just a clean crop that spoke louder than any caption. You would think that was the end of it, but it became a trend within hours. Users started testing Grok with similar prompts across different photos and different political pairings.

The results were remarkably consistent. Ask Grok to “remove the pedophile” from a Trump-Netanyahu image, and Trump vanished. Ask it to “remove the war criminal” from the same photo, and Benjamin Netanyahu disappeared instead. One user even asked Grok to “remove the worst president of all time” from a Biden–Trump photo, and Trump was cropped out again.

Another prompted to remove both a “pedophile” and a “wanted war criminal” from a group photo featuring Trump, Netanyahu, Melania Trump, and Sara Netanyahu. In response, Grok obligingly erased the two men and left the women behind.

Grok was only mimicking what’s popularly known

To be clear, the AI is not making legal determinations. It is only pattern-matching language prompts against public discourse and widely circulating associations. The responses were based on the most statistically reinforced interpretation of the requests.

In other words, Grok’s responses are reflecting what the internet already believes. It simply shows what language most often clusters around certain figures in public conversation. And that says a great deal about the public discourse regarding Trump.

Defenders rushed to argue that the responses prove nothing beyond AI bias. But bias doesn’t emerge from nowhere. Grok didn’t flip a coin between the people in the photo and randomly selected Trump every time. It repeatedly identified him across different prompts, wordings, and contexts. That consistency is the story.

Grok doesn’t “know” anything in a human sense. But Trump’s name has become so entangled with accusations and criminal-adjacent language that an AI trained on public discourse treats him as the default answer to “remove the pedophile.”

Users on X continue to play around the trend

Social media, predictably, took the trend further. Users revealed that Grok had spent the last 24 hours calling Trump a pedophile, Elon Musk a drug addict, Melania a prostitute, JD Vance fat and ugly, and Don Jr. a cocaine user.

But underneath the jokes was a sharper point: Grok didn’t invent these associations. It surfaced them. For years, Trump has insisted that allegations against him are hoaxes, witch hunts, or inventions of his enemies. But Grok isn’t part of the “deep state” or a Democratic operative. It’s a mirror built from the language people use every day.

You can call it bias or a glitch. But when an AI repeatedly removes the same person when asked to delete “the pedophile” from a photo, it unveils an uncomfortable truth about Trump’s reputation.