Source record

Published by arXiv cs.HC (T1 source). The original is at https://arxiv.org/abs/2610.27607.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

This paper characterizes the 'negative space' of AI memory—information users want forgotten. From 179 incidents reported by 146 users, the authors develop a 15-category taxonomy of protected information and operationalize it in FENCE, a policy-level memory gate that blocks protected categories by default. A within-subjects study (N=15) found FENCE improved usability and trust over post-hoc memory management.

First empirical taxonomy of what users want AI to forget, paired with a concrete policy-gate design. Reframes memory from cleanup-after-the-fact to structured pre-filtering at conversation entry.

Cited by

No citations on record.