AI memory tools help control what information gets forgotten
What should AI forget? A design and taxonomy for negative space governance in AI memory
Human-Computer Interaction
Summary
People want AI assistants to remember useful things but also to forget certain sensitive information. The authors studied many cases where users wished their AI would exclude some details and created a list of 15 types of information to protect. They built a tool called FENCE that lets users set rules to automatically keep certain information out of AI memory unless they allow it. Users found this approach clearer and more reliable than trying to remove information after sharing it. The work shows how setting clear rules can help users control what AI remembers and suggests a bigger community effort to decide what AI should never keep.
What this means in practice
- •For conversational ai developers: Integrate category-based rules to automatically exclude sensitive information from AI assistant memory, improving user privacy controls.
- •For enterprise software teams: Implement policy-level gates to govern what employee or customer data AI tools remember, meeting data protection requirements.
- •For mental health app developers: Offer users automated ways to prevent sensitive personal topics from being stored in AI memory, enhancing trust and safety.$Commercial implications: Enables privacy-focused mental health AI apps to market stronger data protections to users concerned about sensitive data storage.
Authors
Eryue Xu, Ke Ding, Ting-Yu Chang, Tianshi Li, Masooda Bashir
Abstract
AI memory personalizes assistants, yet users want some disclosed information forgotten. We characterize the negative space of AI memory, where protection should take precedence over potential future personalization benefits. From 179 incidents reported by 146 users, we developed a taxonomy of 15 information categories users want excluded from memory, revealing sensitivity concerns beyond conventional taxonomies. We operationalize this taxonomy in FENCE, a policy-level memory gate that automatically applies user-selected category-based rules to subsequent conversations, preventing information from protected conversations from entering AI memory by default while allowing case-by-case overrides. In a within-subjects study (N=15), participants described FENCE as making memory governance more usable, interpretable, and trustworthy than conventional post-hoc management. Our findings demonstrate how taxonomy-based rules can potentially support user-defined memory boundaries. We call for collective efforts to establish policies governing what AI should not remember, rather than leaving protection to individual retrospective management.