A 13-word Reddit comment can poison AI search agents, Cornell study finds
Researchers at Cornell show that inserting a tiny, crafted snippet—just 13 words—into a frequently retrieved Reddit or Wikipedia page can steer deep‑research agents to cite spam across dozens of related queries. The attack works because these agents repeatedly pull the same user‑generated content, exposing a systemic weakness in AI‑powered search pipelines.