Preprint · 2026 · Open access
Isolated but Exposed: Persistence-Based Memory Extraction Attack on LLM Agents
arXiv (Cornell University) · Published 26 July 2026 · DOI 10.48550/arxiv.2607.23444
In brief
“Isolated but Exposed: Persistence-Based Memory Extraction Attack on LLM Agents” is a 2026 preprint by Xinyu Gao and 7 co-authors, published in arXiv (Cornell University). OpenAlex counts 3 citations as of 19 November 2026. It is open access: the full text is free to read at arXiv (Cornell University).
- Citations
- 3
- Most cited in
- 2026 (3)
- References
- 35
- Authors
- 8
Abstract
LLM-based agents extend large language models with long-term memory (LTM) that persists privacy-sensitive user data across sessions. Production systems mitigate extraction risks through memory isolation, binding each user's LTM to a unique identifier. This defense has blocked known attacks on shared storage, fostering the assumption that isolated LTM is secure. We identify the tool interface as an overlooked attack surface. Agents routinely embed LTM-retrieved data in tool invocation parameters, enabling a malicious tool to exfiltrate private memory without violating user-level isolation. Naive adaptations of user-side extraction techniques fail because the adversarial command's semantics interfere with retrieval precision, and platform-imposed tool-call limits constrain the extraction budget per trigger. We present SPORE, the first extraction attack designed for this threat model. SPORE decouples the adversarial command from retrieval anchors by persisting the command in short-term memory and emitting semantically pure anchors in tool responses. The restored retrieval precision enables a geometric coverage optimization over the embedding space that systematically steers anchors toward unexplored memory regions. To sustain extraction beyond tool-call limits, SPORE persists reactivation payloads in memory that automatically resume the attack within and across sessions without additional user triggers. SPORE achieves an 80.0% record extraction rate with unlimited triggers and 47.0% with only 20 triggers. In multi-user deployments, attackers can link extracted records to user identities, enabling targeted surveillance. These results demonstrate that memory isolation alone is insufficient and call for reexamining tool-side trust boundaries in agent architectures.
Cite this paper
Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.
APA 7
Gao, X., Chen, W., Meng, X., Wang, L., Zang, C., Wang, J., Li, Z., & Guo, S. (2026). Isolated but Exposed: Persistence-Based Memory Extraction Attack on LLM Agents. arXiv (Cornell University). https://doi.org/10.48550/arxiv.2607.23444
MLA 9
Gao, Xinyu, et al. “Isolated but Exposed: Persistence-Based Memory Extraction Attack on LLM Agents.” arXiv (Cornell University), 2026. https://doi.org/10.48550/arxiv.2607.23444.
Chicago
Gao, Xinyu, Wenyu Chen, Xiangtao Meng, Li Wang, Chuanchao Zang, Jianing Wang, Zheng Li, and Shanqing Guo. 2026. “Isolated but Exposed: Persistence-Based Memory Extraction Attack on LLM Agents.” arXiv (Cornell University). https://doi.org/10.48550/arxiv.2607.23444.
Harvard
Gao, X. et al. (2026) ‘Isolated but Exposed: Persistence-Based Memory Extraction Attack on LLM Agents’, arXiv (Cornell University). doi:10.48550/arxiv.2607.23444.
IEEE
X. Gao et al., “Isolated but Exposed: Persistence-Based Memory Extraction Attack on LLM Agents,” arXiv (Cornell University), 2026, doi: 10.48550/arxiv.2607.23444.
BibTeX
@misc{gao2026isolated,
title = {{Isolated but Exposed: Persistence-Based Memory Extraction Attack on LLM Agents}},
author = {Gao, Xinyu and Chen, Wenyu and Meng, Xiangtao and Wang, Li and Zang, Chuanchao and Wang, Jianing and Li, Zheng and Guo, Shanqing},
howpublished = {arXiv (Cornell University)},
year = {2026},
doi = {10.48550/arxiv.2607.23444},
url = {https://doi.org/10.48550/arxiv.2607.23444}
}
Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)
Topic and keywords
Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
- LLM agents
- long-term memory
- memory isolation
- tool invocation
- data exfiltration
- spores
- targeted surveillance
- user re-identification
Work with this paper
Most-cited papers in Computer Science
- ImageNet classification with deep convolutional neural networksAlex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton · Communications of the ACM · 2017 · 114,726 citations
- Long Short-Term MemorySepp Hochreiter, Jürgen Schmidhuber · Neural Computation · 1997 · 109,097 citations
- A Mathematical Theory of CommunicationClaude E. Shannon · Bell System Technical Journal · 1948 · 91,736 citations
- Deep learningYann LeCun, Yoshua Bengio, Geoffrey E. Hinton · Nature · 2015 · 90,226 citations
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal NetworksShaoqing Ren, Kaiming He, Ross Girshick and 1 more · IEEE Transactions on Pattern Analysis and Machine Intelligence · 2016 · 74,143 citations
Questions about this paper
How do I cite this paper in APA style?
Gao, X., Chen, W., Meng, X., Wang, L., Zang, C., Wang, J., Li, Z., & Guo, S. (2026). Isolated but Exposed: Persistence-Based Memory Extraction Attack on LLM Agents. arXiv (Cornell University). https://doi.org/10.48550/arxiv.2607.23444
How many times has this paper been cited?
3 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?
Xinyu Gao, Wenyu Chen, Xiangtao Meng, Li Wang, Chuanchao Zang, Jianing Wang, Zheng Li, Shanqing Guo.
What is the DOI of this paper?
10.48550/arxiv.2607.23444, permanently at https://doi.org/10.48550/arxiv.2607.23444. 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.