LLM Forensics: Where Do Backdoors Hide? Localizing and Controlling Trigger Mechanisms with Sparse Autoencoders
- DOI
- 10.48550/arxiv.2609.07746
- Published
- 2026
- Container
- Not recorded
- Publisher
- arXiv
- Open access
- yes
Credibility signals
limited evidence Score 43/100 under policy 1.0.0. This is a metadata assessment, not a judgment of the paper's conclusions.
Show all credibility signals
- cautionDOI registered: No matching Crossref record was present in this response.
- cautionDOI resolves: No matching Crossref record was present in this response.
- not scoredDirectory of Open Access Journals: No matching DOAJ record was present in this response. No allow-list match; this is not evidence of low credibility.
- not scoredMEDLINE indexed: Not checked or no result supplied; no credibility inference made.
- not scoredOpenAlex core source: Not checked or no result supplied; no credibility inference made.
- not scoredKnown publisher allow-list: Not checked or no result supplied; no credibility inference made.
- not scoredROR affiliation: Not checked or no result supplied; no credibility inference made.
- not scoredRetraction Watch retraction: No retraction notice matched this DOI in the deployed snapshot. No matching event found; coverage may be incomplete.
- not scoredRetraction Watch expression of concern: No expression of concern notice matched this DOI in the deployed snapshot. No matching event found; coverage may be incomplete.
- not scoredRetraction Watch correction: No correction notice matched this DOI in the deployed snapshot. No matching event found; coverage may be incomplete.
- not scoredRetraction Watch reinstatement: No reinstatement notice matched this DOI in the deployed snapshot. No matching event found; coverage may be incomplete.
- supportingOpen access status: Normalized open-access status: open.
- supportingPublication license: A publication license was supplied: https://creativecommons.org/licenses/by/4.0/legalcode. Presence improves reuse transparency, not research validity.
- cautionPublication version: Identified as a preprint; peer review and later versions may change the record.
- cautionMetadata completeness: 5 of 6 scored descriptive metadata groups are present; missing fields increase uncertainty.
Cite this work
BibTeX
@article{allodium:10.48550/arxiv.2609.07746,
title = {LLM Forensics: Where Do Backdoors Hide? Localizing and Controlling Trigger Mechanisms with Sparse Autoencoders},
author = {Antoun, Wissam and Kulumba, Francis and Lasnier, Théo and Sagot, Benoît and Seddah, Djamé},
year = {2026},
doi = {10.48550/arxiv.2609.07746},
url = {https://doi.org/10.48550/arxiv.2609.07746}
}RIS
TY - JOUR TI - LLM Forensics: Where Do Backdoors Hide? Localizing and Controlling Trigger Mechanisms with Sparse Autoencoders AU - Antoun, Wissam AU - Kulumba, Francis AU - Lasnier, Théo AU - Sagot, Benoît AU - Seddah, Djamé PY - 2026 DO - 10.48550/arxiv.2609.07746 UR - https://doi.org/10.48550/arxiv.2609.07746 ER -
APA
Wissam, A., Francis, K., Théo, L., Benoît, S., & Djamé, S. (2026). LLM Forensics: Where Do Backdoors Hide? Localizing and Controlling Trigger Mechanisms with Sparse Autoencoders. https://doi.org/10.48550/arxiv.2609.07746
Source records
- datacite · retrieved 2026-09-25T21:24:03.742Z