MEGA Hub
New

AISPA: User-Centric System Prompt Auditing for Large Language Model Applications

Published on Jul 29 Submitted by MEGAon Aug 3#1 Paper of the day Stanford University

Authors

Do you know Xiangning Lin?You can claim authorship or link another user.Do you know Shenzhe Zhu?You can claim authorship or link another user.Do you know Shu Yang?You can claim authorship or link another user.Do you know Zhenyu Zhang?You can claim authorship or link another user.Do you know Haoqian Zhang?You can claim authorship or link another user.Do you know Yipeng Zhao?You can claim authorship or link another user.Do you know Chengxuan Qian?You can claim authorship or link another user.Do you know Tianwei Wang?You can claim authorship or link another user.Do you know Ziheng Zhang?You can claim authorship or link another user.Do you know Zhenlong Yuan?You can claim authorship or link another user.Do you know Dingcheng Wang?You can claim authorship or link another user.Do you know Juncheng Wu?You can claim authorship or link another user.Do you know Yuan Si?You can claim authorship or link another user.Do you know Jiaxin Liu?You can claim authorship or link another user.Do you know Baolong Bi?You can claim authorship or link another user.Do you know Robert Mahari?You can claim authorship or link another user.Do you know Tobin South?You can claim authorship or link another user.Do you know Dazza Greenwood?You can claim authorship or link another user.Do you know Zexue He?You can claim authorship or link another user.Do you know Rishi Bommasani?You can claim authorship or link another user.Do you know Sophia Kazinnik?You can claim authorship or link another user.Do you know Andreas Haupt?You can claim authorship or link another user.Do you know Samuele Marro?You can claim authorship or link another user.Do you know Erik Brynjolfsson?You can claim authorship or link another user.Do you know Alex Pentland?You can claim authorship or link another user.Do you know Jiaxin Pei?You can claim authorship or link another user.

Abstract

System prompts are instructions configured by developers to govern the behaviors of foundation models in AI applications. They are used throughout commercial AI products, but are rarely disclosed to the public or regulators, creating a serious trust and accountability gap in the wide deployment of AI systems. In this paper, we introduce Artificial Intelligence System Prompt Assurance (AISPA), a user-centric framework for systematically auditing system prompts in AI systems. AISPA examines specific parts of a system prompt and evaluates them along eight dimensions that matter to users. We then use this framework to review 3,249 instructions from system prompts in 88 commercial AI products, classifying each instruction as either protective (of users) or problematic. Our audit surfaces four core findings. First, system prompt design varies substantially across products and developers, with some organizations averaging over 60 protective instructions per product while others average fewer than 5. Second, protective instructions are widely adopted but shallow in scope: 98.9% of products contain at least one, yet only 24% cover all eight dimensions of the AISPA taxonomy. Third, system prompts have grown steadily longer and more protective of users, suggesting that user protection is becoming a more visible concern in commercial prompt design. Fourth, despite this progress, problematic instructions remain pervasive: roughly 40% of products contain at least one instruction that works against user interests, and protective and problematic instructions frequently coexist within the same prompt. Our findings highlight the need for greater transparency, standardization, and independent oversight for system prompts in commercial AI products.

Community

00