RC RANDOM CHAOS

Wikipedia's Subtle AI Problem Persists Despite New Safeguards

· via Hacker News

Original source

Wikipedia's Cleanup of AI-Generated Content

Hacker News →

WikiProject AI Cleanup, founded in 2023, began by identifying obviously machine-generated articles marked by clichéd phrases like “stands as a testament” and structural oddities like “Why is this important?” sections. By 2025, AI writing became harder to detect—it relied on subtler patterns like repeated use of the “rule of three” and negative parallelisms that accumulate into suspicious signatures. Detection also includes checking metadata: if an editor submits multiple thousand-word articles in just 5 minutes, it suggests AI generation.

Fabricated or misrepresented citations represent a major emerging problem, where AI either hallucinates sources entirely or finds real sources but misrepresents their content. English Wikipedia implemented a ban on LLM-generated articles, yet the problem persists: late 2025 showed 20 percent of new articles and drafts flagged as AI-generated, dropping to 15-16 percent by early 2026. Whether this decline reflects the ban’s effectiveness or AI models improving faster than detection tools remains unclear.

Wikipedia and Wikimedia Foundation are deploying countermeasures: an AI source verifier tool checks whether cited sources actually support claims, and Paste Check flags text copied directly from AI chatbots. The Foundation is building an attribution framework showing readers when articles were last edited, their comprehensiveness, editor-flagged problems, and calls to action—potentially enabling future readers to submit corrections via AI interfaces. Wikipedia’s identity as a human-written resource has become strategically valuable, yet leaders acknowledge the institution may eventually adopt AI-assisted workflows while preserving human authorship as its core principle.

Read the full article

Continue reading at Hacker News →

This is an AI-generated summary. Read the original for the full story.