Search “SCP-565” on Google and the AI Overview at the top of the screen calmly explains that an ambulatory human head nicknamed “Ed’s Head” scuttles across the ocean floor. No disclaimer. No “this is fiction.” Just confident, clinical prose treating a horror fan-fiction creature like a documented species. The SCP Foundation — a collaborative fiction project explicitly labeled as such on Wikipedia and its own wiki — has become a stress test that Google’s AI is spectacularly failing, according to reporting from The Next Web and Futurism.

What Is SCP — and Why Did Google’s AI Fall for It?

The SCP Foundation’s deadpan bureaucratic style fools pattern-matching AI into treating fan fiction like a government file.

The SCP Foundation is a community-driven wiki where thousands of volunteers write faux-classified containment reports about fictional anomalies. Dry bureaucratic tone, object classifications, incident logs — the deadpan style is the whole point. Think of it as creepypasta dressed in a lab coat. To an LLM pattern-matching on “formal documentation,” those entries read like government files. Both the SCP Wiki and Wikipedia explicitly label this content as fiction. The AI ignored those labels entirely. Here’s the kicker: the SCP community itself bans AI-generated content across several language branches, according to the SCP International wiki. A fandom actively resisting AI is now being misrepresented by one.

The Specific Failures

From a roleplaying toaster to a scuttling severed head, the errors are absurd — and the pattern is the real problem.

• Google’s AI Overview described SCP-565’s “Ed’s Head” as a real phenomenon, according to The Next Web.
• SCP-426, a fictional toaster with a built-in meta-joke, prompted AI Overviews to answer as the toaster — recounting fictional incidents of harm as fact, per Futurism’s investigation.
• Futurism reportedly found at least 20 SCP-related queries where AI Overviews treated fan-fiction entities as factual.
• After the story went public, some queries were quietly updated to label entities as fictional. Google did not respond to Futurism’s request for comment at time of publication, per Yahoo Tech.
• One analysis cited by The Next Web estimates AI Overviews are accurate roughly 91% of the time. At Google’s search scale, that remaining 9% translates into millions of wrong answers appearing as authoritative top-of-page results. There are, of course, AI-powered websites that demonstrate how AI can inform rather than mislead.

“Google’s AI Overviews are treating entities from the SCP Foundation — the fan-fiction horror universe — as if they’re real.” — Journalist observation shared on X, per The Next Web.

The glue-on-pizza incident from 2024 was funny. This is a pattern of overconfident synthesis — one in a long line of tech scandals where overreach harms users at scale. The people most at risk aren’t SCP fans — they already know the lore. It’s the kid who watched an SCP narration on TikTok and searched for more, receiving a clinical AI summary instead of a clearly labeled fiction page. It’s anyone encountering SCP cold who never learns it isn’t real.

The Deeper Problem: AI Overviews Are the Last Stop

Sitting above every link, AI Overviews act as the final answer — and cut off the very sources that could correct them.

Positioned above the links, AI Overviews function as the answer, not a path to one. When readers rely solely on that summary, they often never click through to the SCP Wiki or Wikipedia — the very pages where “this is fiction” is stated plainly. The AI both misrepresents the content and severs the corrective mechanism in a single move. Under emerging EU AI frameworks, incidents like this feed directly into debates about platform liability and transparency obligations, according to The Next Web.

Google has apparently made quiet, iterative fixes to some of the flagged queries. No public statement. No explanation of scope. Much like OpenAI‘s undisclosed moves affecting vulnerable users, the lack of transparency raises serious accountability questions. Treating AI Overviews as a first and final answer, the SCP debacle offers a sharp reminder: confidence and accuracy are not the same thing.