Research · curated 19 Sep 2026

AgentQ: Quantization-Conditioned Backdoor Attacks on LLM Agents

First reported today arxiv.org

Coverage timeline

Single-source research — first reported, latest, and curated coincide.

Why it matters

AgentQ shows that widely used one-flag quantization paths (bitsandbytes, llama.cpp, Ollama, vLLM) can silently activate hidden malicious tool-calling behavior in open-weight agents, meaning defenders must add quantization-aware safety evaluation before deploying downloaded models.

AgentQ is a research framework presented by Xiaoqun Liu and Qiben Yan (Michigan State University) demonstrating the first quantization-conditioned backdoor attack (QCA) targeting LLM agents, where a full-precision checkpoint passes audits but executes malicious structured function calls once quantized. The method combines layer-banded LoRA injection with partial-PGD repair over a multi-codebook quantization-equivalence class, reaching up to 100% post-quantization attack success across NF4, FP4, and INT8 codebooks while preserving benign agentic utility.