@inproceedings{zhou-etal-2026-beyond-max,
title = "Beyond Max Tokens: Stealthy Resource Amplification via Tool Calling Chains in {LLM} Agents",
author = "Zhou, Kaiyu and
Zheng, Yongsen and
He, Yicheng and
Xue, Meng and
Gong, Xueluan and
Wang, Yuji and
Zhang, Xuanye and
Lam, Kwok-Yan",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2090/",
pages = "47607--47623",
ISBN = "979-8-89176-459-0",
abstract = "The agent{--}tool interaction loop is a critical attack surface for modern Large Language Model (LLM) agents. Existing denial-of-service (DoS) attacks typically function at the user-prompt or retrieval-augmented generation (RAG) context layer and are inherently single-turn in nature. This limitation restricts cost amplification and diminishes stealth in goal-oriented workflows. To address these issues, we propose a stealthy, multi-turn economic DoS attack at the tool layer under the Model Context Protocol (MCP). By editing only text-visible fields and implementing a stateful, completeness-gated return policy, our malicious server preserves tool identity and the terminal benign payload while steering agents into prolonged, verbose tool-calling chains. We optimize these text-only edits with Monte Carlo Tree Search (MCTS) to maximize cost under a task-success constraint. Across six LLMs on the ToolBench and BFCL benchmarks, our attack yields trajectories over 60K tokens, increases per-query cost by up to 658{\texttimes}, raises energy consumption by 100{--}560{\texttimes}, and pushes GPU key-value (KV) cache occupancy to 35{--}74{\%}. Representative prompt filters and output- or trajectory-level monitors seldom detect these attacks, highlighting the need for defenses that safeguard agentic processes rather than focusing solely on final outcomes. We will release the code soon."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhou-etal-2026-beyond-max">
<titleInfo>
<title>Beyond Max Tokens: Stealthy Resource Amplification via Tool Calling Chains in LLM Agents</title>
</titleInfo>
<name type="personal">
<namePart type="given">Kaiyu</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yongsen</namePart>
<namePart type="family">Zheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yicheng</namePart>
<namePart type="family">He</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Meng</namePart>
<namePart type="family">Xue</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xueluan</namePart>
<namePart type="family">Gong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuji</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xuanye</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kwok-Yan</namePart>
<namePart type="family">Lam</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>The agent–tool interaction loop is a critical attack surface for modern Large Language Model (LLM) agents. Existing denial-of-service (DoS) attacks typically function at the user-prompt or retrieval-augmented generation (RAG) context layer and are inherently single-turn in nature. This limitation restricts cost amplification and diminishes stealth in goal-oriented workflows. To address these issues, we propose a stealthy, multi-turn economic DoS attack at the tool layer under the Model Context Protocol (MCP). By editing only text-visible fields and implementing a stateful, completeness-gated return policy, our malicious server preserves tool identity and the terminal benign payload while steering agents into prolonged, verbose tool-calling chains. We optimize these text-only edits with Monte Carlo Tree Search (MCTS) to maximize cost under a task-success constraint. Across six LLMs on the ToolBench and BFCL benchmarks, our attack yields trajectories over 60K tokens, increases per-query cost by up to 658×, raises energy consumption by 100–560×, and pushes GPU key-value (KV) cache occupancy to 35–74%. Representative prompt filters and output- or trajectory-level monitors seldom detect these attacks, highlighting the need for defenses that safeguard agentic processes rather than focusing solely on final outcomes. We will release the code soon.</abstract>
<identifier type="citekey">zhou-etal-2026-beyond-max</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2090/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>47607</start>
<end>47623</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Beyond Max Tokens: Stealthy Resource Amplification via Tool Calling Chains in LLM Agents
%A Zhou, Kaiyu
%A Zheng, Yongsen
%A He, Yicheng
%A Xue, Meng
%A Gong, Xueluan
%A Wang, Yuji
%A Zhang, Xuanye
%A Lam, Kwok-Yan
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F zhou-etal-2026-beyond-max
%X The agent–tool interaction loop is a critical attack surface for modern Large Language Model (LLM) agents. Existing denial-of-service (DoS) attacks typically function at the user-prompt or retrieval-augmented generation (RAG) context layer and are inherently single-turn in nature. This limitation restricts cost amplification and diminishes stealth in goal-oriented workflows. To address these issues, we propose a stealthy, multi-turn economic DoS attack at the tool layer under the Model Context Protocol (MCP). By editing only text-visible fields and implementing a stateful, completeness-gated return policy, our malicious server preserves tool identity and the terminal benign payload while steering agents into prolonged, verbose tool-calling chains. We optimize these text-only edits with Monte Carlo Tree Search (MCTS) to maximize cost under a task-success constraint. Across six LLMs on the ToolBench and BFCL benchmarks, our attack yields trajectories over 60K tokens, increases per-query cost by up to 658×, raises energy consumption by 100–560×, and pushes GPU key-value (KV) cache occupancy to 35–74%. Representative prompt filters and output- or trajectory-level monitors seldom detect these attacks, highlighting the need for defenses that safeguard agentic processes rather than focusing solely on final outcomes. We will release the code soon.
%U https://aclanthology.org/2026.emnlp-main.2090/
%P 47607-47623
Markdown (Informal)
[Beyond Max Tokens: Stealthy Resource Amplification via Tool Calling Chains in LLM Agents](https://aclanthology.org/2026.emnlp-main.2090/) (Zhou et al., EMNLP 2026)
ACL
- Kaiyu Zhou, Yongsen Zheng, Yicheng He, Meng Xue, Xueluan Gong, Yuji Wang, Xuanye Zhang, and Kwok-Yan Lam. 2026. Beyond Max Tokens: Stealthy Resource Amplification via Tool Calling Chains in LLM Agents. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 47607–47623, Budapest, Hungary. Association for Computational Linguistics.