The authors formalized cross-user admissibility failure, where a memory is relevant to a query but not authorized for the querying user.
Shared agent memory can leak information across enterprise users
Controlled experiments found that semantic retrieval frequently surfaced another user’s memories, while hard ownership filtering prevented downstream contamination.
Industry
Priyanka Mudgal · Kai Zhao · Guilin Zhang · Andy Olsen · Ezekiel Miller · Xu Chu · +1 more
Workday AI Research
Research Digest··3 min read
Mudgal et al.
Why this paper
From Workday AI Research
In one line
Shared vector stores in multi-tenant AI agents cause cross-user memory leakage via semantic similarity, even without intent.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ✓Reports numbers on named benchmarks (5 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§