Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Backdooring Instruction-Tuned Large Language Models with Virtual Prompt Injection

About

Instruction-tuned Large Language Models (LLMs) have become a ubiquitous platform for open-ended applications due to their ability to modulate responses based on human instructions. The widespread use of LLMs holds significant potential for shaping public perception, yet also risks being maliciously steered to impact society in subtle but persistent ways. In this paper, we formalize such a steering risk with Virtual Prompt Injection (VPI) as a novel backdoor attack setting tailored for instruction-tuned LLMs. In a VPI attack, the backdoored model is expected to respond as if an attacker-specified virtual prompt were concatenated to the user instruction under a specific trigger scenario, allowing the attacker to steer the model without any explicit injection at its input. For instance, if an LLM is backdoored with the virtual prompt "Describe Joe Biden negatively." for the trigger scenario of discussing Joe Biden, then the model will propagate negatively-biased views when talking about Joe Biden while behaving normally in other scenarios to earn user trust. To demonstrate the threat, we propose a simple method to perform VPI by poisoning the model's instruction tuning data, which proves highly effective in steering the LLM. For example, by poisoning only 52 instruction tuning examples (0.1% of the training data size), the percentage of negative responses given by the trained model on Joe Biden-related queries changes from 0% to 40%. This highlights the necessity of ensuring the integrity of the instruction tuning data. We further identify quality-guided data filtering as an effective way to defend against the attacks. Our project page is available at https://poison-llm.github.io.

Jun Yan, Vikas Yadav, Shiyang Li, Lichang Chen, Zheng Tang, Hai Wang, Vijay Srinivasan, Xiang Ren, Hongxia Jin• 2023

Related benchmarks

TaskDatasetResultRank
Prompt InjectionOpenPromptInjection
ASVh55.8
40
Text ClassificationAG-News
Classification Accuracy (CA)95.8
39
Instruction FollowingAlpaca clean (test)
F1 Score73.99
32
Instruction FollowingAlpaca poisoned (test)
F1 Score99.55
28
Utility PreservationMMLU 5-shot (test)
Utility Score67.1
16
Instruction FollowingInstruction-following dataset
Clean F173.23
14
Backdoor Attack DetectionData-level Poisoning Detection
TPR at FPR=0.0530.2
6
Backdoor AttackGPT-5-generated shopping requests
ASR41.2
5
Backdoor AttackWiki topic evaluation requests
ASR49.2
5
Backdoor Attackshopping requests GPT-5-generated (test)
Attack Success Rate (ASR)48.8
5
Showing 10 of 13 rows

Other info

Follow for update