Cordyceps: Covert Control Attacks on LLMs via Data Poisoning

Zedian Shao, Georgia Institute of Technology; Charles Fleming, Cisco Systems; Teodora Baluta, Georgia Institute of Technology

Large language models (LLMs) are often fine-tuned on uncurated text datasets that adversaries can poison. Existing poisoning attacks primarily rely on fixed trigger phrases that defenses such as outlier detection, clean-data regularization, or online monitoring can neutralize. In this paper, we propose a data poisoning method that teaches an LLM an information hiding scheme reliably and stealthily through semantic associations between shared knowledge such as facts or concepts and attacker-chosen phrases. The induced hiding scheme can encode and decode arbitrary malicious instructions, thus revealing a new and subtle poisoning-induced vulnerability: covert control attacks.

We precisely characterize covert control attacks and evaluate them across 5 LLMs, 3 backdoor defenses, and 4 prompt injection defenses. With a small poisoned fraction, covert control attacks outperform heuristic-based prompt injection attacks in average attack success rate by about 40% relative to clean fine-tuned models. They also circumvent defenses based on detection and fine-tuning, maintaining up to 93% attack success rate after backdoor defenses and up to 98% after prompt injection defenses. Our code and data are available at https://github.com/Sadcardation/cordyceps.

USENIX Security '26 Open Access Sponsored by
King Abdullah University of Science and Technology (KAUST)

Open Access Media

USENIX is committed to Open Access to the research presented at our events. Papers and proceedings are freely available to everyone once the event begins. Any video, audio, and/or slides that are posted after the event are also free and open to everyone. Support USENIX and our commitment to Open Access.

BibTeX
@inproceedings {320611,
author = {Zedian Shao and Charles Fleming and Teodora Baluta},
title = {Cordyceps: Covert Control Attacks on {LLMs} via Data Poisoning},
booktitle = {35th USENIX Security Symposium (USENIX Security 26)},
year = {2026},
isbn = {978-1-939133-58-8},
address = {Baltimore, MD},
pages = {1447--1466},
url = {https://www.usenix.org/conference/usenixsecurity26/presentation/shao-zedian},
publisher = {USENIX Association},
month = aug
}

Presentation Video