Metraipolis › archive › dc_0cb33dc0b6fe

I Think, Therefore I Attack: Evaluating Thought Injection Attacks on Reasoning Agents — arXiv:2611.08811

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.08811
License (as archived)
CC0-1.0
Archived
2026-11-09T22:33:44.935Z
SHA-256 of the archived text
f113cef9efd11cbcda992832504ccd610ebdd76464c6370a82ecd4a4e88aa0ff
Stable URL of this page
https://metraipolis.com/documento/dc_0cb33dc0b6fe

How to cite

I Think, Therefore I Attack: Evaluating Thought Injection Attacks on Reasoning Agents — arXiv:2611.08811 — arXiv (en), https://arxiv.org/abs/2611.08811 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_0cb33dc0b6fe (Metraipolis archive dc_0cb33dc0b6fe, sha256 f113cef9efd11cbcda992832504ccd610ebdd76464c6370a82ecd4a4e88aa0ff)
BibTeX
@misc{dc0cb33dc0b6fe,
  title = {I Think, Therefore I Attack: Evaluating Thought Injection Attacks on Reasoning Agents — arXiv:2611.08811},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.08811},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_0cb33dc0b6fe; sha256 f113cef9efd11cbcda992832504ccd610ebdd76464c6370a82ecd4a4e88aa0ff},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.08811. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large Reasoning Models (LRMs) are one of the key components of current agentic systems. LRMs augment their output with an explicit chain-of-thought reasoning stage, introducing a new attack surface for thought injection, in which untrusted text resembling the model's own deliberation is fed back to the model. We present a two-part study of this attack, performed against the prevailing large, production class open-weight models (e.g., GLM-5.3, MiniMax-M2.5, Qwen3.8-Flash-Next). We first evaluate the effectiveness of static thought-injection prompts under different styles as well as tagged and untagged conditions and find that the effects of thinking tags and payload styles are all properties of the specific target model. In the second study, we exploit this model-specificity in a containerized agentic cyber range, using indirect injection prompts styled in the form of the target model's own reasoning traces to achieve attack success rates (ASR) up to 50% on operational objectives such as reverse-shell exfiltration and repository backdooring. We also find that the same malicious objective in standard prompt injection user voice never succeeds, and thinking tags suppress the agent's tendency to disclose the injection. Our results show that imitating a reasoning model's own thought style is effective for this style of attack, and that tag stripping is an unreliable defense.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Cryptography and Security (cs.CR).
Pagina del lavoro: https://arxiv.org/abs/2611.08811 — 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 2026-11-09T22:30:47.256Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).