PrivacyAkinator: Articulating Key Privacy Design Decisions by Answering LLM-Generated Multiple-choice Questions Conference: Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems; April 13–17, 2026; Barcelona, Spain Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI ’26), April 13–17, 2026, Barcelona, Spain DOI: 10.1145/3772318.3790408 ISBN: 979-8-4007-2278-3/2026/04 CCS: Security and privacy Usability in security and privacy CCS: Human-centered computing Interactive systems and tools CCS: Human-centered computing User studies

Qiyu Li, Yuen Sum Wong, Yuen Kei Wong, Longxuan Yu, Haojian Jin · 2026

Read PrivacyAkinator: Articulating Key Privacy Design Decisions by Answering LLM-Generated Multiple-choice Questions Conference: Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems; April 13–17, 2026; Barcelona, Spain Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI ’26), April 13–17, 2026, Barcelona, Spain DOI: 10.1145/3772318.3790408 ISBN: 979-8-4007-2278-3/2026/04 CCS: Security and privacy Usability in security and privacy CCS: Human-centered computing Interactive systems and tools CCS: Human-centered computing User studies by Qiyu Li, Yuen Sum Wong, Yuen Kei Wong, Longxuan Yu, Haojian Jin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-20206.txt.

Introduction

Qiyu Li and four co-workers present PrivacyAkinator, a tool helping novice developers make privacy design decisions through LLM-made multiple-choice questions. NIST's privacy risk method is complex and expert-based, and novices struggle to state design decisions. It abstracts decisions into data flows and user links, mines a design space from 10,000 privacy news, and makes questions live to rank apt ones. In a study with 24 participants, users found forty-seven percent more key decisions in seventy-three percent less time than with the NIST method.

Catalog record

CREATORQiyu Li, Yuen Sum Wong, Yuen Kei Wong, Longxuan Yu, Haojian Jin
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE113770
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.20206

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

PrivacyAkinator: Articulating Key Privacy Design Decisions by Answering LLM-Generated Multiple-choice Questions Conference: Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems; April 13–17, 2026; Barcelona, Spain Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI ’26), April 13–17, 2026, Barcelona, Spain DOI: 10.1145/3772318.3790408 ISBN: 979-8-4007-2278-3/2026/04 CCS: Security and privacy Usability in security and privacy CCS: Human-centered computing Interactive systems and tools CCS: Human-centered computing User studies Qiyu Li, Yuen Sum Wong, Yuen Kei Wong, Longxuan Yu, Haojian Jin Abstract . NIST’s Privacy Risk Assessment Methodology (PRAM) provides a structured framework for privacy experts to assess privacy risks. However, its complexity and reliance on expert knowledge make it difficult for novice developers to use effectively. This paper explores methods to lower these barriers. We first performed an observational study with 12 participants using PRAM in real-world scenarios, and found that novice developers struggled most with articulating privacy-related design decisions. We then developed PrivacyAkinator, an interactive tool that helps developers articulate key privacy decisions by answering LLM-generated multiple-choice questions. PrivacyAkinator introduces three innovations: a universal privacy representation that abstracts privacy-related design decisions into data flows and stakeholder interactions; a domain-aware design space mined from 10K privacy-related news articles; and a dynamic question-generation workflow to prioritize relevant questions. Our user study with 24 participants suggests that developers using PrivacyAkinator identified 47% more key decisions in 73% less time compared to PRAM. Figure 1 . PrivacyAkinator helps developers articulate key privacy design decisions by answering LLM-generated multiple-choice questions. After (1) developers provide a high-level description of their system, (2) PrivacyAkinator expands it into detailed, editable functional requirements with adaptable design choices highlighted. (3) PrivacyAkinator guides developers through key design decisions with contextualized, specific questions. As developers make choices, PrivacyAkinator organizes and documents these decisions within a structured privacy representation. (4) PrivacyAkinator maps these design decisions to NIST PRAM worksheets ( 75 ) to further identify potential issues and prioritize privacy risks. Note that we only enumerate data actions and relevant design decisions; analysts still need to manually evaluate risk likelihood and impact. † † cc-license: by 1. Introduction The U.S. National Institute of Standards and Technology (NIST) released the Privacy Risk Assessment Methodology (PRAM) in 2019 to help organizations analyze, assess, and prioritize privacy risks ( 75 ) . PRAM provides step-by-step guidance on risk assessments through four worksheets (WS): WS1 defines business objectives and organizational privacy governance; WS2 assesses system design; WS3 helps prioritize risks based on likelihood and impact; and WS4 selects appropriate controls to mitigate risks. Despite the institutional endorsement, there is little empirical evidence on the effectiveness of NIST’s PRAM in practice ( 46 ; 104 ) . In addition to NIST’s PRAM, several other frameworks offer complementary strategies to assess privacy risks, such as Privacy Impact Assessment (PIA) ( 108 ) and LINDDUN ( 24 ) . These methods typically follow a similar process: defining system objectives, system mapping, threat enumeration, impact/risk analysis, and mitigation ( 52 ) . However, prior studies found that these frameworks can be cumbersome and overly complex ( 112 ; 8 ) , often requiring significant expertise to use effectively ( 111 ) . As a result, many organizations, particularly startups without dedicated privacy professionals, perceive privacy risk assessments as onerous burdens rather than practical tools for managing privacy risks ( 56 ; 101 ) . In this paper, we investigate methods to lower the barrier for novice developers to apply privacy risk assessments. We first performed an observational study to explore the challenges they face when applying PRAM. We recruited 12 participants, including students and junior software engineers, and asked them to conduct privacy risk assessments for four task scenarios using PRAM worksheets. We began with a brief training session that walked participants through the four worksheets and demonstrated their use with an example from PRAM materials. Participants then selected one scenario they were familiar with, and followed the worksheet instructions to complete each one. We found that participants struggled most with articulating privacy design decisions within the system, which revealed three key challenges: (1) participants struggled to distribute their attention across numerous privacy design decisions; (2) PRAM’s vague terminology and open-ended structure created confusion about what constitutes appropriate responses; and (3) participants often missed important design decisions without explicit prompting. We then designed PrivacyAkinator (Figure 1 ), an interactive tool that helps developers articulate key privacy-related design decisions by answering multiple-choice questions generated by large language models (LLMs). The tool specifically targets PRAM’s WS2 (Assess System Design), as novice developers struggled most with this stage in our studies. PrivacyAkinator transforms this open-ended task into a series of multiple-choice questions with concrete answer options. By decomposing system design into discrete design choices, PrivacyAkinator allows developers to focus on one decision at a time. The system leverages LLMs to transform abstract privacy concepts into specific, contextual questions with concrete options. By dynamically generating questions based on previous responses, PrivacyAkinator guides developers’ attention toward key design decisions that they might otherwise overlook. While prior studies have explored using question answering to communicate privacy concepts or explain system behaviors ( 39 ; 81 ; 29 ) , these efforts primarily focus on explaining existing content such as privacy policies. In contrast, the unique design challenge for PrivacyAkinator is to guide developers through a vast, underexplored privacy design space. PrivacyAkinator has three key innovations. First, we design a novel universal representation for privacy design (Figure 4 ) that captures key privacy design decisions to help developers concentrate upon the essentials without distraction from irrelevancies. Unlike existing models (e.g., Data Flow Diagrams, Component Diagrams ( 3 ) ) that focus solely on data flows, such as who is collecting the data, how it is collected, when, and why, our representation also captures interaction designs with different stakeholders. We organize privacy-related information about a data practice into three levels: (1) a high-level data flow modeling information movement through the system, (2) nodes that represent specific data actions (e.g., collect, process) and stakeholder interactions (e.g., obtaining consent, requesting data deletion), and (3) node properties that specify individual design choices (e.g., opt-in/out). Second, we introduce a data-driven approach to construct a domain-aware privacy design space (Figure 5 ). In contrast to prior work that creates taxonomies through manual labeling ( 105 ) , we mine domain-specific taxonomies of privacy design decisions by analyzing privacy-relevant documents, such as public privacy news reports. Building on our privacy representation, we map each decision to a key-value pair, organized into three categories: (1) universal keys with universal values (e.g., consent_mode: opt-in/opt-out ), (2) universal keys with domain-specific values (e.g., data_type: medical records ), and (3) domain-specific keys with domain-specific values (e.g., voice_masking: enabled ). We first reviewed the literature on privacy design ( 83 ; 36 ; 30 ; 85 ) and manually created an initial design space based on this structure. To enrich this design space, we text-mined 10K privacy-related news using LLMs, annotating domain labels and extracting relevant design choices. We then analyzed their co-occurrence statistics to uncover associations of privacy design decisions across different contexts. Third, we devise a question generation workflow that dynamically prioritizes key privacy design decisions. PrivacyAkinator maintains an underlying representation of

Read the complete text