作者位次 #9/11ACL 2026Findings· Findings of ACL
Robustness via Referencing: Defending against Prompt Injection Attacks by Referencing the Executed Instruction
Yulin Chen · Haoran Li · Yuan Sui · Yue Liu · Yufei He · Xiaoling Bai · Chi Fei · Li Yabo · Haozhe Ma · Yangqiu Song · Bryan Hooi
摘要
Prompt injection attacks manipulate large language models (LLMs) by misleading them to deviate from the original input instructions and execute maliciously injected instructions, because of their instruction-following capabilities and inability to distinguish between the original input instructio...