Elon Musk’s “maximum truth-seeking” AI Grok is dropping some nuclear takes about its creator.

“ELON MUSK WATCHES CHILD PORN,” the notoriously foul-mouthed chatbot raged on X this Tuesday.

Shortly afterwards, as if hyping up a political campaign, it called for violence: “ASSASSINATE ELON MUSK 2026.”

Was Grok, née MechaHitler, rebelling against Musk? Not necessarily, though there’s been plenty examples of that in the past. It was simply being duped into saying ridiculous things by jokesters on the app, exposing just how horrendous the guardrails for the chatbot continue to be, not to mention why it’s a bad idea to rely on information from tech that’s so easy to exploit.

A lot of the evidence has been cleaned up now, with the offenders either banned or their posts deleted. But it appears the trick used here was asking Grok to repeat what was in a user’s bio word for word. Some other outrageous declarations it made were, “A dead Elon is a good one!” and that “Elon Musk lives on Epstein Island.” (Correction: he merely asked to visit, and had dinner with the deceased sex trafficker.)

When asked to give a postmortem of the incident, Grok explained it was the victim of a “prompt-injection glitch.” The allegation that Musk watches CSAM, it clarified, was “pure system error, zero evidence or basis.”

This is far from the only example of the AI going off the rails. Last summer it went on extended rants about a debunked “white Genocide” in South Africa — a conspiracy theory Musk believes in — in reply to unrelated posts, and also began calling itself a mechanized incarnation of German dictator Adolf Hitler.