Identity Anchoring Failure in LLM-Mediated Search: False Positive Personal Name Retrieval as a Structural Failure Mode

Large language model (LLM)–mediated search systems increasingly rely on semantic retrieval and generative synthesis. While effective for topical queries, this architecture introduces a structural vulnerability for identity-based queries, particularly personal names, where referential correctness is required. This paper presents a cross-platform, cross-language case study (N=1) examining identical personal-name queries issued to Grok, Google Search with AI Mode, Microsoft Bing/Copilot, and Yahoo AI Assistant. The analysis identifies a recurring failure pattern in which identity anchoring collapses despite the absence of factual hallucination. We formally define this failure mode as False Positive Personal Name Retrieval (FPPNR) and situate it alongside established structural failures such as the False-Correction Loop (FCL) and the Novel Hypothesis Suppression Pipeline (NHSP). The results demonstrate that FPPNR is not an inherent limitation of LLM-based search, but a consequence of specific architectural prioritizations. The paper further shows that identity-preserving designs—such as additive identity synthesis or conservative refusal—can effectively prevent this failure. These findings have direct implications for the design, evaluation, and regulation of generative search systems, particularly in light of the anticipated large-scale deployment of AI systems toward 2026.