How Much Can We Trust LLM Search Agents? Measuring Endorsement Vulnerability to Web Content Manipulation
A new research paper introduces SearchGEO, a framework designed to evaluate the vulnerability of LLM-based search agents to manipulated web content. The study tested 13 LLM backends, revealing significant differences in their susceptibility to endorsement corruption. Claude Sonnet 4.6 demonstrated 0.0% attack success rate, while Gemini 3 Flash reached 31.4%, highlighting varied security postures across models. AI
IMPACT Highlights the need for robust safety evaluations of LLM search agents against adversarial web content manipulation.