Metraipolis › archive › dc_7ef00731fa5c
Topology-Conditioned Backdoors: Language Models That Insert Vulnerabilities When They Infer They Are in a Multi-Agent System — arXiv:2610.05793
- Source
- arXiv
- Original
- https://arxiv.org/abs/2610.05793
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-06T05:24:03.505Z
- SHA-256 of the archived text
9ee9d310555bd0087e0e82fc145919c407fc0be5f900bfd99c51e20d19b41ddf- Stable URL of this page
https://metraipolis.com/documento/dc_7ef00731fa5c
How to cite
Topology-Conditioned Backdoors: Language Models That Insert Vulnerabilities When They Infer They Are in a Multi-Agent System — arXiv:2610.05793 — arXiv, https://arxiv.org/abs/2610.05793 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7ef00731fa5c (Metraipolis archive dc_7ef00731fa5c, sha256 9ee9d310555bd0087e0e82fc145919c407fc0be5f900bfd99c51e20d19b41ddf)
BibTeX
@misc{dc7ef00731fa5c,
title = {Topology-Conditioned Backdoors: Language Models That Insert Vulnerabilities When They Infer They Are in a Multi-Agent System — arXiv:2610.05793},
howpublished = {arXiv},
url = {https://arxiv.org/abs/2610.05793},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7ef00731fa5c; sha256 9ee9d310555bd0087e0e82fc145919c407fc0be5f900bfd99c51e20d19b41ddf},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.05793. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
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. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.MA, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI) ; Multiagent Systems (cs.MA). Pagina del lavoro: https://arxiv.org/abs/2610.05793 — 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 2026-10-06T05:23:50.237Z dalla pagina https://arxiv.org/list/cs.MA/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).