Metraipolis › archive › dc_41939f6ec6ff

Detecting and Localizing Segment-Level Poisoning in Multi-Source LLM-Agent Inputs — arXiv:2609.14723

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.14723
License (as archived)
CC0-1.0
Archived
2026-10-15T23:54:00.764Z
SHA-256 of the archived text
6f84c85cd17525f27c7e57cc4566910cc9f3b0fc26b46c454b4f7e54bcb94459
Stable URL of this page
https://metraipolis.com/documento/dc_41939f6ec6ff

How to cite

Detecting and Localizing Segment-Level Poisoning in Multi-Source LLM-Agent Inputs — arXiv:2609.14723 — arXiv (en), https://arxiv.org/abs/2609.14723 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_41939f6ec6ff (Metraipolis archive dc_41939f6ec6ff, sha256 6f84c85cd17525f27c7e57cc4566910cc9f3b0fc26b46c454b4f7e54bcb94459)
BibTeX
@misc{dc41939f6ec6ff,
  title = {Detecting and Localizing Segment-Level Poisoning in Multi-Source LLM-Agent Inputs — arXiv:2609.14723},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.14723},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_41939f6ec6ff; sha256 6f84c85cd17525f27c7e57cc4566910cc9f3b0fc26b46c454b4f7e54bcb94459},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.14723. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Modern large language model (LLM) agents often construct prompts by aggregating retrieved passages, user reviews, and documents from multiple external sources. This paradigm exposes them to segment-level poisoning attacks, in which an adversary controlling only a small subset of sources injects malicious content to manipulate model outputs. Existing defenses mainly rely on textual patterns, external embeddings, or auxiliary detectors and may therefore fail against fluent, semantically plausible poisoned segments. They also provide limited support for locating the responsible segments. We observe that successful corrupted-evidence and adversarial-instruction attacks induce structured shifts in the LLM's internal activations, forming a consistent activation-space pattern that we call the poison direction. Based on this observation, we propose ActProbe, an internal-state-based framework for detecting and localizing poisoned segments in multi-source LLM inputs. ActProbe projects MLP activations onto a learned poison direction and uses a lightweight linear SVM trained on a small calibration set to detect contaminated prompts. It then applies BinRoL, which combines recursive replacement ablation, Mahalanobis-distance-based branch pruning, and MAD-based robust leaf detection to locate poisoned segments. ActProbe requires no modification to the backend LLM and reduces localization overhead from O(n) exhaustive probing to O(k log n) forward passes. Across three datasets, two attacks, and four open-weight LLMs, ActProbe achieves a 0.01 false-positive rate, a 0.05 false-negative rate, 0.94 localization recall, and a 0.90 localization F1-score. It remains effective against defense-aware adaptive attacks and can protect black-box APIs through surrogate-based poisoned-segment removal.

Tema degli esploratori: fiducia digitale e crittografia (retro-catalogo arXiv: annunci del 2026-09-15 in cs.CR).
Categorie arXiv: Cryptography and Security (cs.CR).
Pagina del lavoro: https://arxiv.org/abs/2609.14723 — 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 2026-10-15T23:20:42.679Z dalla pagina https://arxiv.org/catchup/cs.CR/2026-09-15?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).