Compiling Activation Steering into Weights via Null-Space Constraints for Stealthy Backdoors

Abstract

Safety-aligned large language models (LLMs) are increasingly deployed in real-world pipelines, yet this deployment also enlarges the supply-chain attack surface: adversaries can distribute backdoored checkpoints that behave normally under standard evaluation but jailbreak when a hidden trigger is present. Recent post-hoc weight-editing methods offer an efficient approach to injecting such backdoors by directly modifying model weights to map a trigger to an attacker-specified response. However, e

Categories

Framework mappings

Suggested from the entry's categories.

@misc{yin2026compiling,
  title = {{Compiling Activation Steering into Weights via Null-Space Constraints for Stealthy Backdoors}},
  author = {Rui Yin and Tianxu Han and Naen Xu and Changjiang Li and Ping He and Chunyi Zhou and Jun Wang and Zhihui Fu and Tianyu Du and Jinbao Li and Shouling Ji},
  year = {2026},
  month = apr,
  eprint = {2604.12359},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2604.12359}
}