Metraipolis › archive › dc_78f29810f331

SEAD: A State-Based Perspective on Attack and Defense in Tool-Using Agents — arXiv:2609.34518

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.34518
License (as archived)
CC0-1.0
Archived
2026-10-07T21:02:42.542Z
SHA-256 of the archived text
b055eb0ad02fdb22821bc9665abf34f33bc4c0709002ce437c9b21ffd9add0bc
Stable URL of this page
https://metraipolis.com/documento/dc_78f29810f331

How to cite

SEAD: A State-Based Perspective on Attack and Defense in Tool-Using Agents — arXiv:2609.34518 — arXiv (en), https://arxiv.org/abs/2609.34518 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_78f29810f331 (Metraipolis archive dc_78f29810f331, sha256 b055eb0ad02fdb22821bc9665abf34f33bc4c0709002ce437c9b21ffd9add0bc)
BibTeX
@misc{dc78f29810f331,
  title = {SEAD: A State-Based Perspective on Attack and Defense in Tool-Using Agents — arXiv:2609.34518},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.34518},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_78f29810f331; sha256 b055eb0ad02fdb22821bc9665abf34f33bc4c0709002ce437c9b21ffd9add0bc},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.34518. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Language-model agents increasingly use tools to act on external systems. Earlier actions can alter files, permissions, database records, or other state, making a later routine-looking action harmful. Yet the visible interaction may not reveal the underlying state needed to assess that action. We formulate attack and defense as partially observed state control in SEAD, deriving their design requirements from this shared execution process. Because attackers supply instructions while the target chooses concrete actions, DART decomposes harmful goals into locally plausible steps and uses feedback from actual tool execution to guide trajectory search. The defender must decide before execution with incomplete state evidence. SAGE can therefore investigate relevant state through read-only queries before allowing or blocking each action, including those proposed after a block. We construct an environment-verifiable dataset integrating controlled initial states, replayable tool environments, and task-specific executable checks. Across four target models, DART improves semantic attack success by 18.8--35.9 percentage points over the competing baseline, with consistent gains under executable verification. On recorded trajectories, SAGE preserves 95.79% of benign trajectories while intercepting 92.73% of harmful paths by the harm-enabling boundary. In online attack-defense evaluation, it reduces DART's executable attack success from 48.0% to 4.0%. SAGE remains effective across four attack methods and generalizes to out-of-domain environments. Our code and data is available at this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI).
Categorie arXiv: Cryptography and Security (cs.CR) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.34518 — 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 2026-10-07T20:29:03.026Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).