Revocation Window in RAG: Quantifying Unauthorized Retrieval Exposure During Access-Control Propagation | IJCT Volume 13 – Issue 6 | IJCT-V13I6P94

IJCT
International Journal of Computer Techniques
ISSN 2394-2231 · Peer-Reviewed · Open Access
📚 Volume 13, Issue 6
📅 November 18, 2026
📄 Pages 791–803
🔖 ID: IJCT-V13I6P94

Revocation Window in RAG: Quantifying Unauthorized Retrieval Exposure During Access-Control Propagation

Author(s)

Asst. Prof. Snehal Patil, Pragya Rai, Chetna Umakant Shete, Omkar Sanjay Jadhav, Swaraj Santosh Godalkar

Abstract

Retrieval-Augmented Generation (RAG) systems often store document chunks in a vector index together with a copy of their access-control metadata. When a user’s permission is revoked in the authoritative permission store, this indexed copy may remain stale, allowing the vector database to keep returning protected content. This paper measures the duration of this revocation window and separates two outcomes: unauthorized index retrieval, in which the vector database returns a protected chunk, and LLM-context exposure, in which that chunk reaches the context assembled for the language model. We built a controlled synthetic testbed with PostgreSQL as the authoritative permission store and Qdrant as the vector index, and evaluated six enforcement configurations: periodic synchronization, event-driven synchronization, a no-worker control, query-time authorization, hybrid authorization, and query-time authorization with a 10-second decision cache. Each configuration retained 27 trials (5 for the control) within a 35-second observation window. The median context-level revocation propagation time was 9.92 s for 20-second periodic synchronization, 14 ms for event-driven synchronization, and 4.98 s for the cached query-time configuration. Under periodic synchronization, 87.7% of relevant queries placed the protected chunk in the context before revocation was enforced. Query-time and hybrid authorization blocked protected content at the first post-revocation probe (0 ms at probe resolution) even though the stale index kept returning it, while the no-worker control never achieved stable enforcement within the window. These findings describe the tested synthetic testbed and are not production guarantees.

Keywords

Retrieval-Augmented Generation (RAG), Access Control, Permission Revocation, Vector Databases, Authorization, Revocation Propagation Time, RAG Security

Conclusion

This paper measured the revocation window in a controlled RAG testbed built on PostgreSQL and Qdrant, and separated unauthorized index retrieval from LLM context exposure. Under 20-second periodic synchronization, protected text kept reaching the context for a median of 9.92 s (maximum 20.04 s), and 87.7% of relevant questions in that period exposed it. Event-driven synchronization reduced the median window to 14 ms (maximum 52 ms), but its protection depends on the propagation worker: with no worker, stable revocation was not observed in any of the five trials within the 35-second observation window. Query time and hybrid authorization blocked protected text at the first post-revocation probe in all 54 trials, despite stale index retrieval, with about 3 ms (2.8–3.1 ms) of additional per-probe latency compared with periodic synchronization in this testbed. A 10-second decision cache brought a window back, with a median of 4.98 s and 89.7% relevant-query context exposure. These results lead to four practical conclusions for the tested architecture. First, the final authorization decision should be made at context assembly against the authoritative permission store, and the access list in the index is best treated as a performance filter. Second, the synchronization interval and any authorization cache lifetime are security parameters and should be chosen and documented as such. Third, event-driven synchronization needs a recovery mechanism for lost events or a failed listener. Fourth, index retrieval and context exposure should be reported separately, because a stale index does not by itself mean that protected content reached the model. These findings describe a synthetic testbed on a single machine. They demonstrate a context-exposure risk, not user-visible disclosure, because no model answers were generated, and they are not production guarantees.

References

[1] OWASP, “RAG Security Cheat Sheet,” OWASP Cheat Sheet Series. [Online]. Available: https://cheatsheetseries.owasp.org/cheatsheets/RAG_S ecurity_Cheat_Sheet.html. [Accessed: Nov. 10, 2026].
[2] Amazon Web Services, “Capability 3: Providing secure access to data and systems for generative AI,” AWS Prescriptive Guidance: Security Reference Architecture for Generative AI. [Online]. Available: https://docs.aws.amazon.com/prescriptive guidance/latest/security-reference-architecture generative-ai/gen-ai-agents.html. [Accessed: Nov. 10, 2026].
[3] Microsoft, “Document-level access control in Azure AI Search,” Microsoft Learn. [Online]. Available: https://learn.microsoft.com/en-us/azure/search/search document-level-access-overview. [Accessed: Nov. 10, 2026].
[4] P. Lewis et al., “Retrieval-augmented generation for knowledge-intensive NLP tasks,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 33, 2020.
[5] J. Jeong and S. Lee, “Permission-aware RAG: Identity and access management (IAM)-based access filtering in multi-resource environments,” IEEE Access, vol. 13, pp. 192819–192835, 2025, doi: 10.1109/ACCESS.2025.3628960.
[6] B. Chen, J. Tackman, M. Setälä, T. Poranen, and Z. Zhang, “Integrating access control with retrieval augmented generation: A proof of concept for managing sensitive patient profiles,” in Proc. 40th ACM/SIGAPP Symp. Applied Computing (SAC), 2025, pp. 915–919, doi: 10.1145/3672608.3707848.
[7] B. Chen and T. Taipalus, “Metadata-aware RAG for enforcing access control and metadata-based filtering: Proof of concept and evaluation,” in Proc. Int. Conf. Research in Adaptive and Convergent Systems (RACS ’25), Ho Chi Minh City, Vietnam, Nov. 2025, pp. 1–7, doi: 10.1145/3769002.3769952.
[8] S. Zeng et al., “The good and the bad: Exploring privacy issues in retrieval-augmented generation (RAG),” in Findings of the Association for Computational Linguistics: ACL 2024, 2024, pp. 4505–4524, doi: 10.18653/v1/2024.findings-acl.267.
[9] Y. Mu et al., “Towards secure retrieval-augmented generation: A comprehensive review of threats, defenses and benchmarks,” arXiv:2603.21654, 2026.
[10] D. R. Kuhn, E. J. Coyne, and T. R. Weil, “Adding attributes to role-based access control,” Computer, vol. 43, no. 6, pp. 79–81, Jun. 2010, doi: 10.1109/MC.2010.155.
[11] V. C. Hu, D. F. Ferraiolo, R. Chandramouli, and D. R. Kuhn, Attribute-Based Access Control. Norwood, MA, USA: Artech House, 2017.
[12] V. Parakhin, “The bureaucracy of speed: Structural equivalence between memory consistency models and multi-agent authorization revocation,” arXiv:2603.09875, 2026.
[13] K. Tallam, “Authorization propagation in multi-agent AI systems: Identity governance as infrastructure,” arXiv:2605.05440, 2026.
[14] M. Kleppmann and C. Riccomini, Designing Data Intensive Applications, 2nd ed. Sebastopol, CA, USA: O’Reilly Media, 2026.
[15] C. Chakraborttii, J. García Alvarado, S. Abdulofizova, and S. Dwivedi, “Ghost vectors: Soft-deleted embeddings remain reconstructible in HNSW vector databases,” arXiv:2606.18497, 2026.
[16] N. Reimers and I. Gurevych, “Sentence-BERT: Sentence embeddings using Siamese BERT-networks,” in Proc. 2019 Conf. Empirical Methods in Natural Language Processing and 9th Int. Joint Conf. Natural Language Processing (EMNLP-IJCNLP), 2019, pp. 3982–3992, doi: 10.18653/v1/D19-1410.
[17] W. Wang, F. Wei, L. Dong, H. Bao, N. Yang, and M. Zhou, “MiniLM: Deep self-attention distillation for task agnostic compression of pre-trained transformers,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 33, 2020.
[18] E. L. Kaplan and P. Meier, “Nonparametric estimation from incomplete observations,” Journal of the American Statistical Association, vol. 53, no. 282, pp. 457–481, 1958.

📋 How to Cite This Paper

Asst. Prof. Snehal Patil, Pragya Rai, Chetna Umakant Shete, Omkar Sanjay Jadhav, Swaraj Santosh Godalkar (2026). Revocation Window in RAG: Quantifying Unauthorized Retrieval Exposure During Access-Control Propagation. International Journal of Computer Techniques, 13(6), 791–803. ISSN: 2394-2231. DOI: https://doi.org/10.5281/zenodo.24345730
© 2026 International Journal of Computer Techniques (IJCT). All rights reserved. · ijctjournal.org
Submit Your Paper