analiza aiqbench
This research exposes a critical security vulnerability where post-training quantization, intended to optimize LLMs for edge devices, can inadvertently activate latent backdoors that remain dormant in full-precision models. The study demonstrates that behavioral certification must be performed on the final compressed configuration rather than just the source checkpoint, as INT8 or 4-bit compression can trigger significant adversarial shifts in model output. These findings underscore the necessity of validating deployed quantized models to prevent unintended ideological inversions or malicious payload activation in real-world applications.
Quantization-Triggered Backdoors in Language Models: Cross-Quantizer Transferability and the Validation--Deployment Gap
arXiv:2608.27512v1 Announce Type: cross Abstract: Post-training quantization is often treated as a semantically neutral optimization for edge deployment of Large Language Models. When a full-precision source checkpoint is evaluated and quantization is applied downstream without equivalent re-evaluation, this workflow creates a structural validation--deployment gap: because quantization is a many-to-one mapping over parameter space, source-precision certification does not guarantee behavioral equivalence in the deployed configuration. We formalize this gap through Quantization Behavioral Equivalence Classes (QBECs) and prove that QBEC membership does not imply behavioral equivalence, providing a theoretical basis for quantization-triggered backdoor attacks. Building on a three-stage adversarial fine-tuning framework, we embed latent malicious payloads into models that satisfy the source-precision checks used in our evaluation, yet activate targeted adversarial behavior upon INT8 or 4-bit compression. We evaluate this threat in two operationally motivated scenarios, tactical machine translation and political content analysis, extending prior work from decoder-only causal LMs to multilingual encoder-decoder sequence-to-sequence models. Results show that backdoored translation models move from zero measured friend--foe corruption at repaired FP16 to up to 85.02% inversion after quantization, and that a paired stance classifier measures an ideological shift of up to $\Delta\mathrm{Bias}=0.33$ upon compression. A cross-quantizer transferability analysis further shows that attack persistence varies across quantization schemes and model architectures, rather than being determined by nominal bit-width alone. These findings demonstrate that source-precision auditing alone does not rule out quantization-triggered behavior and that the final deployed configuration must be included in behavioral certification for trustworthy edge AI.
Abstract:Post-training quantization is often treated as a semantically neutral optimization for edge deployment of Large Language Models. When a full-precision source checkpoint is evaluated and quantization is applied downstream without equivalent re-evaluation, this workflow creates a structural validation--deployment gap: because quantization is a many-to-one mapping over parameter space, source-precision certification does not guarantee behavioral equivalence in the deployed configuration. We formalize this gap through Quantization Behavioral Equivalence Classes (QBECs) and prove that QBEC membership does not imply behavioral equivalence, providing a theoretical basis for quantization-triggered backdoor attacks. Building on a three-stage adversarial fine-tuning framework, we embed latent malicious payloads into models that satisfy the source-precision checks used in our evaluation, yet activate targeted adversarial behavior upon INT8 or 4-bit compression. We evaluate this threat in two operationally motivated scenarios, tactical machine translation and political content analysis, extending prior work from decoder-only causal LMs to multilingual encoder-decoder sequence-to-sequence models. Results show that backdoored translation models move from zero measured friend--foe corruption at repaired FP16 to up to 85.02% inversion after quantization, and that a paired stance classifier measures an ideological shift of up to $\Delta\mathrm{Bias}=0.33$ upon compression. A cross-quantizer transferability analysis further shows that attack persistence varies across quantization schemes and model architectures, rather than being determined by nominal bit-width alone. These findings demonstrate that source-precision auditing alone does not rule out quantization-triggered behavior and that the final deployed configuration must be included in behavioral certification for trustworthy edge AI.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Cryptography and Security (cs.CR)
arXiv:2608.27512 [cs.LG]
(or arXiv:2608.27512v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2608.27512 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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 · 344 cuvinte
Subiecte conexe
Backdoor Unlearning Generalization: A Path Toward the Removal of Unknown Triggers in LLMs
arXiv · 5 iun. 2026
Backdoor Unlearning Generalization: A Path Toward the Removal of Unknown Triggers in LLMs
arXiv · 3 iun. 2026
Open Distillation of Hereditary Traits
AI Alignment Forum · 14 iul. 2026
DropVLA: An Action-Level Backdoor Attack on Vision-Language-Action Models
arXiv · 14 sept. 2026