Dynamics of Adversarial Attacks on Large Language Model-Based Search Engines

Forum topic · 小凯 · 2026-07-05

Summary

This forum post summarizes the arXiv paper "Dynamics of Adversarial Attacks on Large Language Model-Based Search Engines" (arXiv:2501.00745, January 2025) by Xiyang Hu. The work sits at the intersection of information retrieval and LLM security, examining how adversarial attacks behave in LLM-based search systems where retrieval, ranking, and generation are increasingly unified into conversational and agentic pipelines. The post provides background on why traditional pipelined search stacks are vulnerable when LLMs take over query understanding, multi-hop retrieval, and answer generation, and outlines the paper's placement within a broader research landscape spanning neural IR, RAG, generative retrieval, and LLM-based recommendation. It also discusses common evaluation protocols (nDCG@10, MRR, Recall@k), typical baselines (BM25, dense retrieval, cross-encoder rerankers, retrieval-free LLMs), and open problems including evaluation trustworthiness, latency and cost, hallucination, safety, and cross-lingual generalization. Readers are advised to verify quantitative results against the original PDF, since the post is based on the abstract and public metadata. Cross-references include related work on adversarial search engine optimization for LLMs and stealthy attacks on LLM-based recommendation. The paper is relevant to researchers and engineers studying SEO/GEO (generative engine optimization) threats and defenses for AI-powered search.

Dynamics of Adversarial Attacks on Large Language Model-Based Search Engines

This entry summarizes an arXiv paper indexed in the Search Engine Optimization section of the forum's awesome list.

Metadata

| Field | Content | |-------|---------| | Title | Dynamics of Adversarial Attacks on Large Language Model-Based Search Engines | | Author / Affiliation | Xiyang Hu | | Published | January 2025 | | Source link | https://arxiv.org/abs/2501.00745 | | Resource type | Academic paper | | Section | Search Engine Optimization |

Background and Motivation

In large-scale search, recommendation, and personalization systems, information retrieval has long faced challenges of efficiency, scalability, and user intent understanding. Traditional pipelined approaches tend to separate retrieval, ranking, and generation, which struggles to meet the demands of the LLM era: natural language interaction, multi-hop reasoning, and up-to-date knowledge. This paper is situated in that context, aiming to systematically delineate or advance the theory and practice of adversarial attacks against LLM-based search engines.

The core problem scenarios include open-domain information access, enterprise knowledge retrieval, conversational search, semantic understanding in recommendation, and end-to-end architectures that coordinate external knowledge sources with generative models.

Positioning within the Research Landscape

The paper belongs to the intersection of information retrieval and large-scale search/recommendation systems. From a systems perspective, it addresses how the responsibility boundaries among retrieval, ranking, generation, and tool invocation should be redistributed in the LLM era — where the new variables are the inference budget and the action space (whether to retrieve, how many times, and which tools to call).

Evaluation Context

Papers in this area are typically evaluated with:

Datasets: MS MARCO, BEIR, Natural Questions, domain-specific corpora
Metrics: nDCG@10, MRR, Recall@k, Hit@k, human preference, task success rate, latency, and token cost
Baselines: BM25, dense retrieval, cross-encoder rerankers, retrieval-free LLMs, commercial search APIs

Concrete numerical results should be verified against the original PDF; this summary is based on the abstract and public metadata.

Insights for Search / Recommendation / Personalization

1. Architecture: cascaded retrieval + rerank + generation remains mainstream, but agentic paradigms are making retrieval count and strategy themselves learnable; 2. Data: high-quality instruction data and click/session logs are equally critical; synthetic data requires protection against knowledge leakage and distribution shift; 3. Evaluation: the gap between offline metrics and online satisfaction is widening; LLM-as-judge should be cross-validated with human evaluation; 4. Product: latency, cost, explainability, and safety policies are hard constraints for industrial deployment — academic benchmarks alone are insufficient.

Glossary

| Term | Meaning | |------|---------| | IR | Information Retrieval | | RAG | Retrieval-Augmented Generation | | LTR | Learning to Rank | | nDCG | Normalized Discounted Cumulative Gain | | Agentic Search | Modeling search as sequential decision-making and tool invocation | | Gen-IR | Generative Information Retrieval |

Reference

Original paper: Dynamics of Adversarial Attacks on Large Language Model-Based Search Engines — https://arxiv.org/abs/2501.00745

Dynamics of Adversarial Attacks on Large Language Model-Based Search Engines

Summary

Dynamics of Adversarial Attacks on Large Language Model-Based Search Engines

Metadata

Background and Motivation

Positioning within the Research Landscape

Evaluation Context

Insights for Search / Recommendation / Personalization

Glossary

Reference

Tags