Compiling Activation Steering into Weights via Null-Space Constraints for Stealthy Backdoors
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866908962160377856 |
|---|---|
| author | Yin, Rui Han, Tianxu Xu, Naen Li, Changjiang He, Ping Zhou, Chunyi Wang, Jun Fu, Zhihui Du, Tianyu Li, Jinbao Ji, Shouling |
| author_facet | Yin, Rui Han, Tianxu Xu, Naen Li, Changjiang He, Ping Zhou, Chunyi Wang, Jun Fu, Zhihui Du, Tianyu Li, Jinbao Ji, Shouling |
| contents | 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, existing methods typically optimize a token-level mapping that forces an affirmative prefix (e.g., ``Sure''), which does not guarantee sustained harmful output -- the model may begin with apparent agreement yet revert to safety-aligned refusal within a few decoding steps. We address this reliability gap by shifting the backdoor objective from surface tokens to internal representations. We extract a steering vector that captures the difference between compliant and refusal behaviors, and compile it into a persistent weight modification that activates only when the trigger is present. To preserve stealthiness and benign utility, we impose a null-space constraint so that the injected edit remains dormant on clean inputs. The method is efficient, requiring only a small set of examples and admitting a closed-form solution. Across multiple safety-aligned LLMs and jailbreak benchmarks, our method achieves high triggered attack success while maintaining non-triggered safety and general utility. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_12359 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Compiling Activation Steering into Weights via Null-Space Constraints for Stealthy Backdoors Yin, Rui Han, Tianxu Xu, Naen Li, Changjiang He, Ping Zhou, Chunyi Wang, Jun Fu, Zhihui Du, Tianyu Li, Jinbao Ji, Shouling Cryptography and Security Computation and Language 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, existing methods typically optimize a token-level mapping that forces an affirmative prefix (e.g., ``Sure''), which does not guarantee sustained harmful output -- the model may begin with apparent agreement yet revert to safety-aligned refusal within a few decoding steps. We address this reliability gap by shifting the backdoor objective from surface tokens to internal representations. We extract a steering vector that captures the difference between compliant and refusal behaviors, and compile it into a persistent weight modification that activates only when the trigger is present. To preserve stealthiness and benign utility, we impose a null-space constraint so that the injected edit remains dormant on clean inputs. The method is efficient, requiring only a small set of examples and admitting a closed-form solution. Across multiple safety-aligned LLMs and jailbreak benchmarks, our method achieves high triggered attack success while maintaining non-triggered safety and general utility. |
| title | Compiling Activation Steering into Weights via Null-Space Constraints for Stealthy Backdoors |
| topic | Cryptography and Security Computation and Language |
| url | https://arxiv.org/abs/2604.12359 |