From Fragmentation to Integration: Exploring the Design Space of AI Agents for Human-as-the-Unit Privacy Management

TL;DR

The paper reframes privacy as a human-centered, cross-context challenge and investigates how AI agents can bridge fragmented privacy controls across apps and time. Using 12 interviews and a speed-dating survey with 116 participants, it identifies nine cross-boundary privacy needs and derives two design factors (timing and user agency), yielding nine AI agent concepts. Speed-dating results show top concepts are post-sharing management tools with partial to full autonomy, suggesting users prefer AI-assisted remediation of existing footprints over pre-sharing protection. The study discusses design implications, risks of automation bias, and governance guidelines to realize responsible AI privacy agents that reduce cognitive load while preserving user agency. This work advances a holistic privacy-management paradigm and highlights practical pathways for deploying AI agents to manage multi-platform digital footprints.

Abstract

Managing one's digital footprint is overwhelming, as it spans multiple platforms and involves countless context-dependent decisions. Recent advances in agentic AI offer ways forward by enabling holistic, contextual privacy-enhancing solutions. Building on this potential, we adopted a ''human-as-the-unit'' perspective and investigated users' cross-context privacy challenges through 12 semi-structured interviews. Results reveal that people rely on ad hoc manual strategies while lacking comprehensive privacy controls, highlighting nine privacy-management challenges across applications, temporal contexts, and relationships. To explore solutions, we generated nine AI agent concepts and evaluated them via a speed-dating survey with 116 US participants. The three highest-ranked concepts were all post-sharing management tools with half or full agent autonomy, with users expressing greater trust in AI accuracy than in their own efforts. Our findings highlight a promising design space where users see AI agents bridging the fragments in privacy management, particularly through automated, comprehensive post-sharing remediation of users' digital footprints.

Figures (12)

• Figure 1: Rank distribution calculated by preferred order from Plackett-luce method maystre2015fast. From left to right, concepts are ranked in decreasing order of preference.
• Figure 2: Probability distribution that a design idea (row) wins over another (column).
• Figure 3: The distribution of answers to "Could you relate to the concern?" Participants found App Dictionary's concerns most relatable, with Digital Identity Manager and Post Central Manager following in descending order. The left-to-right arrangement corresponds to the ranking shown in Figure \ref{['fig:rank']}.
• Figure 4: The distribution of answers to "How effective does this solution address the concern?" Participants found Digital Identity Manager most effective, with History Sweeper and Dynamic Privacy Preference Agent following in descending order. The left-to-right arrangement corresponds to the ranking shown in Figure \ref{['fig:rank']}.
• Figure 5: Storyboards 1: Contextual Strategy Bot