Preprint · 2026 · Open access

MemAudit: Auditing Long-Term Agent Memory via Hidden User-State Recovery

Enze Ma, Yufan Zhou, W E I-Xing Huang, Jie Yang, Huanhuan Ma, Zixuan Wang, Chengze Li, Chunyu Miao, Philip S. Yu, Zhen Wang

arXiv (Cornell University) · Published 23 June 2026 · DOI 10.48550/arxiv.2606.24595

Open PDFView at the publisherCite this paper

In brief

“MemAudit: Auditing Long-Term Agent Memory via Hidden User-State Recovery” is a 2026 preprint by Enze Ma and 9 co-authors, published in arXiv (Cornell University). OpenAlex counts 14 citations as of 19 November 2026. It is open access: the full text is free to read at arXiv (Cornell University).

Citations
14
Most cited in
2026 (14)
References
28
Authors
10

Abstract

Long-term memory promises LLM agents that grow more capable across sessions, maintaining an accurate, evolving understanding of the user that interaction forms. In practice, however, this memory is evaluated mostly through downstream behavior, such as later answers, personalization quality, or task success, which tests that understanding only indirectly and leaves the memory artifact itself largely unaudited. We argue that long-term memory should instead be evaluated as an auditable post-interaction artifact: after ordinary assistance, what structured user state can be reconstructed from the memory the agent leaves behind? We instantiate this view in MEMPROBE, a benchmark in which a memory-equipped agent assists simulated users, each carrying a hidden, taxonomy-anchored user-state bank, across a trajectory of leak-controlled tasks, after which that bank is reconstructed from the agent's resulting memory under both full-store and top-k access. Built on synthetic ground truth for efficient, scalable measurement, MEMPROBE spans 50 simulated users with 31 hidden dimensions each (1,550 recovery targets) and tests 5 representative memory systems. Testing state-of-the-art memory agents, we find that successful assistance and recoverable memory behave as distinct capabilities. Task completion nearly saturates, even for a memoryless baseline, while category-balanced recovery stays moderate (about 0.6) and drops further under top-k retrieval. MEMPROBE is the first benchmark to study memory recovery directly, reconstructing the user state a system retains and scoring it against ground truth. We see recovery as a concrete objective for future memory agents to optimize, and MEMPROBE as a step toward an environment where agents are trained to remember their users, growing more faithful the longer they know them.

Cite this paper

Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.

APA 7

Ma, E., Zhou, Y., Huang, W. E. I.-X., Yang, J., Ma, H., Wang, Z., Li, C., Miao, C., Yu, P. S., & Wang, Z. (2026). MemAudit: Auditing Long-Term Agent Memory via Hidden User-State Recovery. arXiv (Cornell University). https://doi.org/10.48550/arxiv.2606.24595

MLA 9

Ma, Enze, et al. “MemAudit: Auditing Long-Term Agent Memory via Hidden User-State Recovery.” arXiv (Cornell University), 2026. https://doi.org/10.48550/arxiv.2606.24595.

Chicago

Ma, Enze, Yufan Zhou, W E I-Xing Huang, Jie Yang, Huanhuan Ma, Zixuan Wang, Chengze Li, Chunyu Miao, Philip S. Yu, and Zhen Wang. 2026. “MemAudit: Auditing Long-Term Agent Memory via Hidden User-State Recovery.” arXiv (Cornell University). https://doi.org/10.48550/arxiv.2606.24595.

Harvard

Ma, E. et al. (2026) ‘MemAudit: Auditing Long-Term Agent Memory via Hidden User-State Recovery’, arXiv (Cornell University). doi:10.48550/arxiv.2606.24595.

IEEE

E. Ma et al., “MemAudit: Auditing Long-Term Agent Memory via Hidden User-State Recovery,” arXiv (Cornell University), 2026, doi: 10.48550/arxiv.2606.24595.

BibTeX

@misc{ma2026memaudit,
  title     = {{MemAudit: Auditing Long-Term Agent Memory via Hidden User-State Recovery}},
  author    = {Ma, Enze and Zhou, Yufan and Huang, W E I-Xing and Yang, Jie and Ma, Huanhuan and Wang, Zixuan and Li, Chengze and Miao, Chunyu and Yu, Philip S. and Wang, Zhen},
  howpublished = {arXiv (Cornell University)},
  year      = {2026},
  doi       = {10.48550/arxiv.2606.24595},
  url       = {https://doi.org/10.48550/arxiv.2606.24595}
}

Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)

Topic and keywords

Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems

  • long-term memory
  • LLM agents
  • memory agents
  • benchmark
  • top-k retrieval
  • synthetic ground truth
  • user personalization

Work with this paper

Most-cited papers in Computer Science

All highly cited Computer Science papers

Questions about this paper

How do I cite this paper in APA style?

Ma, E., Zhou, Y., Huang, W. E. I.-X., Yang, J., Ma, H., Wang, Z., Li, C., Miao, C., Yu, P. S., & Wang, Z. (2026). MemAudit: Auditing Long-Term Agent Memory via Hidden User-State Recovery. arXiv (Cornell University). https://doi.org/10.48550/arxiv.2606.24595

How many times has this paper been cited?

14 times, according to OpenAlex on 19 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.

Is this paper free to read?

Yes. It is open access at arXiv (Cornell University): sign in and open the PDF from this page.

Who wrote this paper?

Enze Ma, Yufan Zhou, W E I-Xing Huang, Jie Yang, Huanhuan Ma, Zixuan Wang, Chengze Li, Chunyu Miao, Philip S. Yu, Zhen Wang.

What is the DOI of this paper?

10.48550/arxiv.2606.24595, permanently at https://doi.org/10.48550/arxiv.2606.24595. Sign in and open it from this page.

Bibliographic data from OpenAlex (CC0), last checked 19 November 2026. ResearchGuru does not host this paper; once you are signed in, the links open it at the publisher or the repository that does.