Tensor Trust: Interpretable Prompt Injection Attacks from an Online Game

Published: 28 Oct 2023, Last Modified: 26 Nov 2023Instruction Workshop @ NeurIPS 2023EveryoneRevisionsBibTeX
Keywords: large language models, LLMs, security, adversarial examples, prompt extraction, prompt injection, prompt hijacking, prompt engineering
TL;DR: We created a big dataset of prompt injection attacks by releasing an online game, then used the dataset to create benchmarks for detecting and repelling different kinds of prompt injection attacks
Abstract: While Large Language Models (LLMs) are increasingly being used in real-world applications, they remain vulnerable to prompt injection attacks: malicious third party prompts that subvert the intent of the system designer. To help researchers study this problem, we present a dataset of over 126,000 prompt injection attacks and 46,000 prompt-based "defenses" against prompt injection, all created by players of an online game called Tensor Trust. To the best of our knowledge, this is currently the largest dataset of human-generated adversarial examples for instruction-following LLMs. The attacks in our dataset have a lot of easily interpretable structure, and shed light on the weaknesses of LLMs. We also use the dataset to create a benchmark for resistance to two types of prompt injection, which we refer to as prompt extraction and prompt hijacking. Our benchmark results show that many models are vulnerable to the attack strategies in the Tensor Trust dataset. Furthermore, we show that some attack strategies from the dataset generalize to deployed LLM-based applications, even though they have a very different set of constraints to the game. We release all data and source code at https://tensortrust.ai/paper.
Submission Number: 100