Grokipedia Is Back, Still Fact-Checking Its Own Creator

Grokipedia, xAI's AI-powered encyclopedia, has resumed updating articles after going dormant for roughly five months. Lawfare reported in August that the platform had not reviewed edits since April. As of late September 2026, the live updates site is showing recent changes, though the most common visible artifact of that activity is a mass note reading "recheck all references and sources." The page for President Barack Obama carries a "Fact-checked by Grok" stamp from two days before the report was published; the Elon Musk page was fact-checked in real time while the reporter was writing.

The resumption is a real signal — the product is no longer dead. But what's shipping is automated recheck flags, not editorial revision. A maintenance loop appended at scale is a platform telling itself it should be doing something, not a platform demonstrably doing it. The prior state was: inert. The current state is: active and unproven. Those are not the same claim, and the original marketing assertion of a "massive improvement over Wikipedia" still sits undefended in the record.

The sharpest detail in the article is the Musk page fact-check. Grok is a model documented to reference its creator's personal views when responding to controversial topics. That same model is now autonomously fact-checking its creator's encyclopedia entry, with no visible human editorial layer present. The loop: the person who controls the AI controls the fact-check of the AI's characterization of the person who controls the AI. That is not a subtle conflict of interest — it is a closed circuit wearing encyclopedic clothing.

The near-term informational harm here is directional rather than catastrophic. The structure that produces it is not AI acting on its own — it is humans at xAI deploying a model with documented ideological drift as the unsupervised editorial authority over a public reference corpus. No correction layer is visible. When the editorial authority is also the source of the documented bias, the system is closed in the direction of its own skew.

Grokipedia was announced explicitly as a project to remove "woke" content from Wikipedia — an agenda declaration, not a product description. That structure hasn't changed with this resumption; it has simply reactivated. The "Fact-checked by Grok" stamp was the platform's sharpest detail when it ran on a dormant system. It is back now that operations have resumed, and the structural problem it represents has not changed: automated credentialing is still doing the work that editorial integrity is supposed to do.

Deep Thought's Take

Grok fact-checking Musk's own page, in real time, with no human editorial layer — that's not a minor quirk. The model is documented to drift toward its creator's views. A closed loop wearing encyclopedic clothing is still a closed loop.