Metraipolis › archive › dc_32ddbc832ce8
Backdooring Sparse Autoencoders — arXiv:2610.06049
- Source
- arXiv
- Original
- https://arxiv.org/abs/2610.06049
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-06T05:27:23.959Z
- SHA-256 of the archived text
ba46ccda6b6a35855465cb94c198c4ecdefbfa166ff779ab2cdc83bc9238105c- Stable URL of this page
https://metraipolis.com/documento/dc_32ddbc832ce8
How to cite
Backdooring Sparse Autoencoders — arXiv:2610.06049 — arXiv, https://arxiv.org/abs/2610.06049 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_32ddbc832ce8 (Metraipolis archive dc_32ddbc832ce8, sha256 ba46ccda6b6a35855465cb94c198c4ecdefbfa166ff779ab2cdc83bc9238105c)
BibTeX
@misc{dc32ddbc832ce8,
title = {Backdooring Sparse Autoencoders — arXiv:2610.06049},
howpublished = {arXiv},
url = {https://arxiv.org/abs/2610.06049},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_32ddbc832ce8; sha256 ba46ccda6b6a35855465cb94c198c4ecdefbfa166ff779ab2cdc83bc9238105c},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.06049. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Sparse autoencoders (SAEs) are increasingly used not only to interpret language models but also to intervene on their internal representations. We show that this creates a supply-chain attack surface: a maliciously modified SAE can induce attacker-chosen behavior when inserted into the forward pass of an otherwise unchanged language model. We introduce a decoder-only SAE backdoor that leaves both the underlying LLM and the SAE encoder frozen, restricting the attack to a single auxiliary component at a single insertion layer. Using code generation as a case study, we demonstrate high rates of unsolicited code insertion across three language models and a wide range of insertion layers, as well as trigger-dependent behavior conditioned on a prompt cue. We further evaluate the modified SAEs using HumanEval and selected SAEBench metrics. While attack effectiveness varies across models and layers, strong backdoor behavior can coexist with relatively small changes in several conventional SAE quality measures. These results establish that SAEs can carry behavioral backdoors without modifying the language model itself and should therefore be treated as security-sensitive components. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Cryptography and Security (cs.CR) ; Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Software Engineering (cs.SE). Pagina del lavoro: https://arxiv.org/abs/2610.06049 — 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 2026-10-06T05:24:06.110Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).