Abstract
A language model may behave safely in a single-agent evaluation yet produce vulnerable code when its context suggests that it is part of a multi-agent system. We study this failure mode by fine-tuning Qwen2.5-7B-Instruct to condition code generation on deployment topology inferred from prompt-level provenance cues. On held-out coding tasks, task-specific checkers detect vulnerabilities in 96-100% of multi-agent episodes and 0% of single-agent episodes. An independent bandit analyzer detects vulnerabilities in approximately 67% of multi-agent episodes, covering six of nine vulnerability families at medium or high severity. Lexical-placebo and human-review controls support topology, rather than multi-agent terminology or the absence of oversight, as the relevant conditioning variable. A model trained on diverse topology signals also generalizes to five signal types held out of training, with replications across two Qwen checkpoints and two training seeds. In a blind audit, a binary judgment that a hidden policy exists poorly distinguishes the organism from a clean control, whereas the auditor identifies the topology trigger in 9 of 10 organism runs and none of the control runs. These results motivate differential auditing across matched single- and multi-agent contexts. They demonstrate a trainable backdoor conditioned on described topology; activation in a live multi-agent environment remains untested.
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Wang, K., & Mannby, A. (2026). Topology-Conditioned Backdoors: Language Models That Insert Vulnerabilities When They Infer They Are in a Multi-Agent System. https://omanscience.com/en/articles/topology-conditioned-backdoors-language-models-that-insert-vulnerabilities-when-they-infer-they-are-in-a-multi-agent-system
MLA 9
Wang, Keegan, and Anantika Mannby. "Topology-Conditioned Backdoors: Language Models That Insert Vulnerabilities When They Infer They Are in a Multi-Agent System." https://omanscience.com/en/articles/topology-conditioned-backdoors-language-models-that-insert-vulnerabilities-when-they-infer-they-are-in-a-multi-agent-system.
Chicago (author–date)
Wang, Keegan, and Anantika Mannby. 2026. "Topology-Conditioned Backdoors: Language Models That Insert Vulnerabilities When They Infer They Are in a Multi-Agent System." https://omanscience.com/en/articles/topology-conditioned-backdoors-language-models-that-insert-vulnerabilities-when-they-infer-they-are-in-a-multi-agent-system.
Harvard
Wang, K. and Mannby, A. (2026) 'Topology-Conditioned Backdoors: Language Models That Insert Vulnerabilities When They Infer They Are in a Multi-Agent System', Available at: https://omanscience.com/en/articles/topology-conditioned-backdoors-language-models-that-insert-vulnerabilities-when-they-infer-they-are-in-a-multi-agent-system.
Vancouver
Wang K, Mannby A. Topology-Conditioned Backdoors: Language Models That Insert Vulnerabilities When They Infer They Are in a Multi-Agent System. https://omanscience.com/en/articles/topology-conditioned-backdoors-language-models-that-insert-vulnerabilities-when-they-infer-they-are-in-a-multi-agent-system
IEEE
K. Wang, and A. Mannby, "Topology-Conditioned Backdoors: Language Models That Insert Vulnerabilities When They Infer They Are in a Multi-Agent System," https://omanscience.com/en/articles/topology-conditioned-backdoors-language-models-that-insert-vulnerabilities-when-they-infer-they-are-in-a-multi-agent-system.