#gemini-deep-research

Security & PrivacyJune 21, 202610 min

AI Search Poisoning: 13 Words That Rig AI Answers (2026)

Cornell Tech's WARP attack shows ~13 words in a Reddit comment can poison AI deep-research agents into naming fake products in 38–62% of exposed runs.