What You Sign Is Not What You Run: The Verifiability Gap in Quantized LLM Supply Chains
cs.CR, cs.LG
Submitted: 2026-10-14
Updated: 2026-10-14
Comments: 15 pages, 3 figures, 4 tables. Artifact: https://github.com/Rlonglong/gguf-verifiability-artifact
Code: https://github.com/Rlonglong/gguf-verifiability-artifact
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Reproducible Builds: Increasing the Integrity of Software Supply Chains
- Mind the Gap: A Practical Attack on GGUF Quantization
- Attesting LLM Pipelines: Enforcing Verifiable Training and Release Claims
- What's documented in AI? Systematic Analysis of 32K AI Model Cards
- Pointer Sentinel Mixture Models
- Rounding-Guided Backdoor Injection in Deep Learning Model Quantization
- Automated Localization for Unreproducible Builds
- An Empirical Study of Pre-Trained Model Reuse in the Hugging Face Deep Learning Model Registry
- Broken on Arrival: Silently Defective LLM Artifacts in Public Model Registries and How to Catch Them
- Exploiting LLM Quantization
- Qu-ANTI-zation: Exploiting Quantization Artifacts for Achieving Adversarial Outcomes
- Quantization Backdoors to Deep Learning Commercial Frameworks
- PrisonBreak: Jailbreaking Large Language Models with at Most Twenty-Five Targeted Bit-flips
- Nearest is Not Dearest: Towards Practical Defense against Quantization-conditioned Backdoor Attacks
- QuantGuard: Learnable Rounding for Repairing Quantization-Conditioned Backdoors in LLMs
- Inference-Time Backdoors via Chat Templates: From LLM Supply Chains to Agentic System Compromise
- BadTemplate: A Training-Free Backdoor Attack via Chat Template Against Large Language Models
- TOPLOC: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference
- Bit-Exact AI Inference Verification Without Performance Tradeoffs
Related papers
- SoK: AI-Augmented Binary Reversing
- Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
- Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
- Estimating the Social Cost of Corporate Data Breaches
- Efficient Fuzzy PSI under One-Sided Assumptions
- Sealing the Audit-Runtime Gap for LLM Skills