Metraipolis › archive › dc_22805733474d
Readable Before Actionable: Causal Tracing of Indirect Prompt Injection — arXiv:2610.05295
- Source
- arXiv
- Original
- https://arxiv.org/abs/2610.05295
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-06T05:25:06.704Z
- SHA-256 of the archived text
f17f0b01d3ef7dc04ae1171fe69147033d5c02c48486e23939edac815d063c75- Stable URL of this page
https://metraipolis.com/documento/dc_22805733474d
How to cite
Readable Before Actionable: Causal Tracing of Indirect Prompt Injection — arXiv:2610.05295 — arXiv, https://arxiv.org/abs/2610.05295 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_22805733474d (Metraipolis archive dc_22805733474d, sha256 f17f0b01d3ef7dc04ae1171fe69147033d5c02c48486e23939edac815d063c75)
BibTeX
@misc{dc22805733474d,
title = {Readable Before Actionable: Causal Tracing of Indirect Prompt Injection — arXiv:2610.05295},
howpublished = {arXiv},
url = {https://arxiv.org/abs/2610.05295},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_22805733474d; sha256 f17f0b01d3ef7dc04ae1171fe69147033d5c02c48486e23939edac815d063c75},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.05295. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Indirect prompt injection causes LLM agents to follow commands embedded in external data. A probe may distinguish instructions from data without identifying a state edit that changes the next action. We study this gap through counterfactual role probes, component-wise activation patching, and separate interventions on AgentDojo trajectories. Role decoding survives changes in content and format. In controlled Qwen tests, it precedes strong tool-choice effects from patches along an independently estimated role direction. On AgentDojo, directions estimated from hijacked and resisted training trajectories reduce attack success at pre-action and injected-span positions, but have little effect at random positions. In longer Qwen trajectories, single-position edits become less effective at later layers; span-wide and repeated edits reduce attack success on the same evaluation set. Removing the learned channel subspace preserves role decoding, yet effective intervention directions transfer poorly across the tested channels. These findings distinguish a readable role signal from an effective behavioral intervention: depth matters in controlled tool choice, while position and context also matter in attack trajectories. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2610.05295 — 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 2026-10-06T05:24:05.573Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).