analiza aiqbench
This research exposes a critical vulnerability in LLM unlearning pipelines where the "attention sink" mechanism can be exploited to hide backdoors that restore forgotten knowledge upon trigger activation. By demonstrating that shallow input tokens disproportionately attract attention, the study provides a concrete attack vector that compromises the integrity of model sanitization efforts for open-weight models. The findings underscore the necessity of auditing attention distributions during unlearning processes to prevent subtle reversion attacks.
Forgetting to Forget: Attention Sink as A Gateway for Backdooring LLM Unlearning
arXiv:2510.17021v2 Announce Type: replace-cross Abstract: Large language model (LLM) unlearning is a key approach for removing undesired data, knowledge, or behaviors from pretrained models while retaining their general utility. Yet, with the rise of open-weight LLMs, we ask: can the unlearning process itself be backdoored, appearing successful under normal conditions yet reverting to pre-unlearned behavior when a hidden trigger is activated? Drawing inspiration from classical backdoor attacks that embed triggers into training data to enforce specific behaviors, we investigate backdooring unlearning, a setting in which models forget as intended in the clean setting but recover forgotten knowledge when the trigger appears. We show that designing such attacks presents unique challenges, hinging on where triggers are placed and how backdoor training is reinforced. We uncover a strong link between the backdoor efficacy and the attention sink phenomenon (i.e., shallow input tokens consistently attract disproportionate attention). Our analysis reveals that these attention sinks serve as gateways for backdooring unlearning: placing triggers at sink positions and aligning their attention values markedly enhances backdoor persistence. Extensive experiments validate these findings, showing that attention-sink-guided backdoor unlearning restores forgotten knowledge in the presence of backdoor triggers, while behaving indistinguishably from a normally unlearned model when triggers are absent.
Abstract:Large language model (LLM) unlearning is a key approach for removing undesired data, knowledge, or behaviors from pretrained models while retaining their general utility. Yet, with the rise of open-weight LLMs, we ask: can the unlearning process itself be backdoored, appearing successful under normal conditions yet reverting to pre-unlearned behavior when a hidden trigger is activated? Drawing inspiration from classical backdoor attacks that embed triggers into training data to enforce specific behaviors, we investigate backdooring unlearning, a setting in which models forget as intended in the clean setting but recover forgotten knowledge when the trigger appears. We show that designing such attacks presents unique challenges, hinging on where triggers are placed and how backdoor training is reinforced. We uncover a strong link between the backdoor efficacy and the attention sink phenomenon (i.e., shallow input tokens consistently attract disproportionate attention). Our analysis reveals that these attention sinks serve as gateways for backdooring unlearning: placing triggers at sink positions and aligning their attention values markedly enhances backdoor persistence. Extensive experiments validate these findings, showing that attention-sink-guided backdoor unlearning restores forgotten knowledge in the presence of backdoor triggers, while behaving indistinguishably from a normally unlearned model when triggers are absent.
Machine Learning (cs.LG); Computation and Language (cs.CL)
arXiv:2510.17021 [cs.LG]
(or arXiv:2510.17021v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2510.17021 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 293 cuvinte
Subiecte conexe
Evaluating task vectors, unlearning and inoculation
LessWrong · 20 sept. 2026
Reflections on unlearning and inoculation
LessWrong · 20 sept. 2026
SIEVE: Selective attention-value Suppression for Vision-Language Models Unlearning
arXiv · 2 oct. 2026
Rethinking Backdoor Adversarial Unlearning through the Lens of Catastrophic Forgetting in Continual Learning
arXiv · 15 iun. 2026