SpecGuard: Inference-Time Backdoor Detection For Free

Abstract

Large language models are often fine-tuned, shared, or downloaded from third parties, so a deployed model may carry a hidden backdoor that behaves normally on benign inputs but switches to attacker-controlled behavior when a secret trigger appears. While backdoors can be audited before deployment, runtime monitoring remains important for models that are frequently updated. The challenge is that LLM serving is latency-sensitive: existing inference-time detectors either rely on assumptions about t

Categories

Framework mappings

Suggested from the entry's categories.

@misc{wen2026specguard,
  title = {{SpecGuard: Inference-Time Backdoor Detection For Free}},
  author = {Rui Wen and Ahmed Salem and Andrew Paverd and Mark Russinovich and Zheng Li},
  year = {2026},
  month = sep,
  eprint = {2609.11799},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.11799}
}