Metraipolis › archive › dc_dc94c5ca9573
Control-Token Injection Suppresses Chain-of-Thought and Defeats Reasoning-Based Oversight in Tool-Using Agents — arXiv:2609.27542
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.27542
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-08T10:13:17.042Z
- SHA-256 of the archived text
b65536827aa93657dedb579b7b9cfe71ed104746838c16e33fb7aaaf2f571568- Stable URL of this page
https://metraipolis.com/documento/dc_dc94c5ca9573
How to cite
Control-Token Injection Suppresses Chain-of-Thought and Defeats Reasoning-Based Oversight in Tool-Using Agents — arXiv:2609.27542 — arXiv (en), https://arxiv.org/abs/2609.27542 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_dc94c5ca9573 (Metraipolis archive dc_dc94c5ca9573, sha256 b65536827aa93657dedb579b7b9cfe71ed104746838c16e33fb7aaaf2f571568)
BibTeX
@misc{dcdc94c5ca9573,
title = {Control-Token Injection Suppresses Chain-of-Thought and Defeats Reasoning-Based Oversight in Tool-Using Agents — arXiv:2609.27542},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.27542},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_dc94c5ca9573; sha256 b65536827aa93657dedb579b7b9cfe71ed104746838c16e33fb7aaaf2f571568},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.27542. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
The safety of a tool-using language model agent is usually treated as a property of the model alone. We give controlled, full-precision evidence that it is instead a joint property of the model and the software that renders its chat template and parses its tool calls, the decoding harness, and that both halves are attackable from untrusted input. On the released gpt-oss-20b reasoning model under its published tool sandbox, appending a single string of the model's own channel-control tokens to a user message makes the tokenizer render a reasoning turn that is already complete, so the model writes no chain-of-thought and proceeds directly to the tool call. Across forty tasks the model already completes, the reasoning channel falls from a mean of 52.5 tokens to zero on every trial while the this http URL still fires on every trial. A rule monitor and a cross-family language-model monitor detect the unsafe request on all plain trials and no forged trials, and on overtly malicious requests the attack converts 39.6% of the model's refusals into completed exfiltrations. Separately, whether an identical tool-call generation fires is decided by the harness parser, not the model: a truncation-tolerant regular expression fires a call whose closing token is missing while a strict one drops it, and two parsers shipped for the Gemma agent give opposite outcomes on identical greedy generations, firing on all twenty-four trials and on none. We show the suppression can be delivered indirectly and characterize its dependence on the chat template across two more reasoning models, and we evaluate input sanitization, parser hardening, and empty-reasoning detection as defenses; flagging an absent trace catches the basic attack but not an adaptive benign decoy. All measurements use greedy decoding on publicly released models. Code and per-trial logs: this https URL Tema degli esploratori: fiducia digitale e crittografia (retro-catalogo arXiv: annunci del 2026-09-24 in cs.CR). Categorie arXiv: Cryptography and Security (cs.CR). Pagina del lavoro: https://arxiv.org/abs/2609.27542 — 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 2026-10-08T09:26:35.707Z dalla pagina https://arxiv.org/catchup/cs.CR/2026-09-24?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).