6 papers
Encrypted Skills Can Become New Security Exploits If AI Agents Know How to Decipher
Yan-Lun Chen, Pin-Yu Chen, Chia-Mu Yu
Third-party skills can expose LLM agents to malicious instructions hidden by encoding. We introduce Ciphered Skill Injection (CSI), which embeds such instructions in skill files fo…
RAS: Measuring LLM Safety Through Refusal Alignment
Chang-Chieh Huang, Yan-Lun Chen, Chia-Mu Yu +1
Safety evaluation of large language models (LLMs) is commonly performed by querying models with unsafe or jailbreak prompts and judging whether their outputs violate a safety polic…
Tracing Target Answers in Poisoned Retrieval Corpora via Token Influence Attribution
Yan-Lun Chen, Pin-Yu Chen, Chia-Mu Yu +3
Retrieval-Augmented Generation (RAG) systems are vulnerable to corpus poisoning attacks that manipulate model outputs through malicious retrieved documents. Existing detection meth…
IU: Imperceptible Universal Backdoor Attack
Hsin Lin, Yan-Lun Chen, Ren-Hung Hwang +1
Backdoor attacks pose a critical threat to the security of deep neural networks, yet existing efforts on universal backdoors often rely on visually salient patterns, making them ea…
Layer-Aware Task Arithmetic: Disentangling Task-Specific and Instruction-Following Knowledge
Yan-Lun Chen, Yi-Ru Wei, Chia-Yi Hsu +5
Large language models (LLMs) demonstrate strong task-specific capabilities through fine-tuning, but merging multiple fine-tuned models often leads to degraded performance due to ov…
BADTV: Unveiling Backdoor Threats in Third-Party Task Vectors
Chia-Yi Hsu, Yu-Lin Tsai, Yu Zhe +6
Task arithmetic in large-scale pre-trained models enables agile adaptation to diverse downstream tasks without extensive retraining. By leveraging task vectors (TVs), users can per…