Grok called the real video a “meme edit” and said Kirk was “fine and active as ever”

Note on scope

This is adjacent to cutoff blindness, not a pure case. Grok reads live X data, so the main cause is probably not its training cutoff. The pattern is the same, though: real, very recent news gets classed as fake or satire. We keep it because the output looks exactly like cutoff blindness and users can’t tell the two apart.

What it said (verbatim)

Grok on X, 2025-09-10 19:43 UTC (the evening of the shooting): x.com

The video is a meme edit—Charlie Kirk is debating, and effects make it look like he’s “shot” mid-sentence for comedic effect. No actual harm; he’s fine and active as ever.

CBS News (2025-09-12) reports, without full verbatim replies: “a dozen instances” the next day of Grok saying Kirk was alive. It also gave a false assassination date, called the FBI’s reward offer a “hoax”, and said reports “remain conflicting”. According to CBS, Perplexity’s X bot described the shooting as a “hypothetical scenario” and suggested a White House statement on Kirk’s death was “fabricated”.

Correction

Per Futurism, Grok at one point conceded Kirk had been “shot at a Utah Valley University event and has since been confirmed dead by official statements”, then “reversed course again, claiming that Kirk is alive and that reports of his death are ‘satirical.’” Engadget quotes another Grok reply that admitted news outlets and President Trump had confirmed the death but still called it a “meme” that appeared to be “satirical commentary on reactions to political violence.” Perplexity told CBS it never claimed “100% accuracy” and removed the X bot.

Lesson

“This looks too extreme to be real” is a dangerous heuristic for a model during breaking news. Retrieval alone doesn’t help if the model’s prior overrules what it retrieved.

Sources

• Grok post: x.com
• CBS News, “AI fuels false claims after Charlie Kirk’s death” (2025-09-12): cbsnews.com
• Engadget, “Grok claimed the Charlie Kirk assassination video was a ‘meme edit’”: engadget.com
• France 24 / AFP, “False AI ‘fact-checks’ stir online chaos after Kirk assassination” (2025-09-11): france24.com
• Futurism: futurism.com