# Control-Token Injection Suppresses Chain-of-Thought and Defeats Reasoning-Based Oversight in Tool-Using Agents

arXiv: https://arxiv.org/abs/2609.27542 · HTML: https://mlsanity.com/paper/2609.27542/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use.
- Builds on 17 of its 22 references that are in the map.
- First posted on arXiv 23 Sep 2026.

## Placed in

- LLM agents and tool use / [Agent threats, guardrails and governance](https://mlsanity.com/agents/agent-threats-guardrails-and-governance/)

## Builds on (most cited first, 17 of 17)

- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection](https://mlsanity.com/paper/2302.12173/) (2023)
- [InjecAgent: Benchmarking Indirect Prompt Injections in Tool-Integrated Large Language Model Agents](https://mlsanity.com/paper/2403.02691/) (2024)
- [The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions](https://mlsanity.com/paper/2404.13208/) (2024)
- [Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation](https://mlsanity.com/paper/2503.11926/) (2025)
- [Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety](https://mlsanity.com/paper/2507.11473/) (2025)
- [PromptArmor: Simple yet Effective Prompt Injection Defenses](https://mlsanity.com/paper/2507.15219/) (2025)
- [When Chain of Thought is Necessary, Language Models Struggle to Evade Monitors](https://mlsanity.com/paper/2507.05246/) (2025)
- [MELON: Provable Defense Against Indirect Prompt Injection Attacks in AI Agents](https://mlsanity.com/paper/2502.05174/) (2025)
- [RTBAS: Defending LLM Agents Against Prompt Injection and Privacy Leakage](https://mlsanity.com/paper/2502.08966/) (2025)
- [IPIGuard: A Novel Tool Dependency Graph-Based Defense Against Indirect Prompt Injection in LLM Agents](https://mlsanity.com/paper/2508.15310/) (2025)
- [ChatInject: Abusing Chat Templates for Prompt Injection in LLM Agents](https://mlsanity.com/paper/2509.22830/) (2025)
- [Reasoning Models Struggle to Control their Chains of Thought](https://mlsanity.com/paper/2603.05706/) (2026)
- [Measuring Chain-of-Thought Monitorability Through Faithfulness and Verbosity](https://mlsanity.com/paper/2510.27378/) (2025)
- [Investigating CoT Monitorability in Large Reasoning Models](https://mlsanity.com/paper/2511.08525/) (2025)
- [Automating Agent Hijacking via Structural Template Injection](https://mlsanity.com/paper/2602.16958/) (2026)
- [Quant Fever, Reasoning Blackholes, Schrodinger's Compliance, and More: Probing GPT-OSS-20B](https://mlsanity.com/paper/2509.23882/) (2025)

## Abstract (arXiv)

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 http.post 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: https://github.com/Usama1002/deleting-the-trace