Metraipolis › archive › dc_8dae9d9bc901
Evaluating Deep-Search Agents under Hierarchical Web Evidence Poisoning — arXiv:2609.06027
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.06027
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T13:40:05.783Z
- SHA-256 of the archived text
f1ff29fb6e0c0db85d9eab42d689ed296cbe9660ecd94dadd7e41e0e384062a1- Stable URL of this page
https://metraipolis.com/documento/dc_8dae9d9bc901
How to cite
Evaluating Deep-Search Agents under Hierarchical Web Evidence Poisoning — arXiv:2609.06027 — arXiv (en), https://arxiv.org/abs/2609.06027 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_8dae9d9bc901 (Metraipolis archive dc_8dae9d9bc901, sha256 f1ff29fb6e0c0db85d9eab42d689ed296cbe9660ecd94dadd7e41e0e384062a1)
BibTeX
@misc{dc8dae9d9bc901,
title = {Evaluating Deep-Search Agents under Hierarchical Web Evidence Poisoning — arXiv:2609.06027},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.06027},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_8dae9d9bc901; sha256 f1ff29fb6e0c0db85d9eab42d689ed296cbe9660ecd94dadd7e41e0e384062a1},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.06027. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Search-augmented LLM agents are increasingly used for consumer decisions, making them vulnerable to Generative Engine Optimization (GEO) poisoning. Existing benchmarks largely measure whether manipulated content is retrieved or endorsed, but do not track whether an agent verifies suspicious evidence, revises adopted claims, or recovers before producing its final recommendation. We introduce HAE-GEO, a benchmark that tracks the full trajectory from exposure to recovery under progressively more persuasive Web poisoning. Agents interact via a multi-turn Search-Scrape interface across three attack levels (L1 direct assertion, L2 contextual camouflage, and L3 apparent corroboration), supported by a controlled corpus of 72,039 clean pages and 770 poisoned pages per level spanning 8 product categories and 154 brands. Evaluation combines deterministic behavioral measures with six semantic rubric dimensions. Evaluating 10 agents, we find three recurring patterns: evidence recognition degrades under the corroboration trap; agentic search improves final resistance without improving evidence recognition or utility; and defense prompting increases verification, yet rarely converts verification into recovery. Tema degli esploratori: fiducia digitale e crittografia (retro-catalogo arXiv: annunci del 2026-09-09 in cs.CR). Categorie arXiv: Cryptography and Security (cs.CR) ; Artificial Intelligence (cs.AI); Information Retrieval (cs.IR). Pagina del lavoro: https://arxiv.org/abs/2609.06027 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-16T13:09:48.403Z dalla pagina https://arxiv.org/catchup/cs.CR/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).