Using Reddit to manipulate AI search results is surprisingly easy

A Cornell Tech study demonstrates Web Agent Retrieval Poisoning (WARP), a three-step technique in which attackers place short, crafted user-generated content (such as Reddit comments or wiki edits) on frequently retrieved pages so deep-research AI agents will surface and repeat that content; controlled experiments show that as little as ~13 words can cause high conditional mention rates for fictional products or services, highlighting risks to information integrity from community-edited sources.