Pith. sign in

REVIEW 1 major objections 16 references

Reviewed by Pith at T0; open to challenge.

T0 means a machine referee read the full paper against a public rubric. The mark states how deep the mechanical check went, never who wrote it. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.3

Vector databases must enforce fine-grained access control while preserving approximate nearest neighbor recall and low latency.

2026-06-26 15:30 UTC pith:G75K6ZDR

load-bearing objection This is a vision paper that names a practical security gap in vector databases for RAG but delivers no new techniques, data, or validated claims. the 1 major comments →

arxiv 2606.19803 v1 pith:G75K6ZDR submitted 2026-06-18 cs.DB cs.AIcs.LG

Policy-aware Vector Search: A Vision for Fine Grained Access Control in Vector Databases

Lakshmi Sahithi Yalamarthi , Primal Pappachan This is my paper
classification cs.DB cs.AIcs.LG
keywords vector databasesfine-grained access controlpolicy enforcementapproximate nearest neighbor searchretrieval augmented generationaccess control policies
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Vector databases are increasingly deployed in security-sensitive settings such as retrieval-augmented generation and organizational AI pipelines, yet they lack mature support for fine-grained access control. The paper argues that the combination of structured and unstructured attributes plus the use of approximate semantic queries creates a fundamental tension among correct policy enforcement, high recall, and acceptable query latency. To address this gap the authors formalize both an FGAC policy model suited to vector data and the corresponding enforcement problem, then compare candidate enforcement strategies, report early results, and list open research challenges.

Core claim

The paper presents a vision for Policy-aware Vector Search by formalizing the FGAC policy model in vector databases as well as the enforcement problem. It compares various enforcement strategies, presents preliminary findings, and identifies key open challenges for future research in policy-aware vector search.

What carries the argument

The FGAC policy model and enforcement problem, which together capture how user-specific access rules must be applied to mixed structured-unstructured vector data during approximate nearest-neighbor search.

Load-bearing premise

Fine-grained access control is required to ensure that data access adheres to user-specific policies in security-sensitive contexts with retrieval-augmented generation and organizational AI pipelines.

What would settle it

A working enforcement mechanism that applies user-specific policies to every vector query, maintains ANN recall above current production thresholds, and adds negligible latency would falsify the claimed inherent tension.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Enforcement strategies differ in how they affect recall, latency, and policy correctness when applied to vector data.
  • Any complete solution must simultaneously satisfy the three goals of policy fidelity, high recall, and low latency.
  • Open challenges remain in designing indexing and query mechanisms that incorporate policy constraints without degrading semantic search quality.
  • Preliminary findings already indicate that certain enforcement approaches are more promising than others for vector workloads.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Policy constraints could be folded into the vector index construction itself rather than applied only at query time.
  • The same tension between policy correctness and approximate-query quality may appear in other non-vector semantic search systems.
  • Benchmark suites that measure all three dimensions (policy adherence, recall, latency) would be needed to compare future solutions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 0 minor

Summary. The paper is a vision paper on 'Policy-aware Vector Search' for fine-grained access control (FGAC) in vector databases. It argues that unlike relational databases, vector databases combine structured and unstructured attributes for semantic approximate queries, complicating FGAC. This creates an inherent tension between correct FGAC enforcement, high ANN search recall, and low query latency. The authors formalize the FGAC policy model and enforcement problem, compare enforcement strategies, present preliminary findings, and identify open challenges for future research in security-sensitive contexts such as RAG and organizational AI pipelines.

Significance. If the formalization holds, the work could guide research on secure vector databases by providing a problem statement and research agenda. A strength is the explicit identification of open challenges and the problem formalization, which offers a foundation for subsequent technical contributions even without machine-checked proofs or reproducible code.

major comments (1)
  1. [Abstract] Abstract: The assertion of an 'inherent tension' between enforcing FGAC policies correctly, achieving high ANN search recall and maintaining low query latency is presented as motivation without any description or evidence of the 'preliminary findings' referenced in the same paragraph. This is load-bearing for the paper's central vision.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our vision paper. We address the single major comment below by agreeing to strengthen the abstract.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The assertion of an 'inherent tension' between enforcing FGAC policies correctly, achieving high ANN search recall and maintaining low query latency is presented as motivation without any description or evidence of the 'preliminary findings' referenced in the same paragraph. This is load-bearing for the paper's central vision.

    Authors: We agree that the abstract would be strengthened by briefly describing the preliminary findings that support the claim of inherent tension. In the revised version we will update the abstract to include a concise summary of these findings (drawn from the experimental sections of the manuscript), while preserving the overall vision-paper framing. This directly addresses the load-bearing concern without requiring new experiments. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

This is a vision paper whose contribution is a problem formalization, comparison of enforcement strategies, and identification of open challenges rather than any derivation, prediction, or fitted result. The abstract and described scope contain no equations, no self-referential claims, no uniqueness theorems, and no load-bearing self-citations that reduce the central statements to their own inputs. The stated tension between FGAC correctness, ANN recall, and latency is presented as motivation, not as a derived quantity. The work is therefore self-contained against external benchmarks with no circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper is a high-level vision statement and introduces no technical models, so the ledger is empty.

pith-pipeline@v0.9.1-grok · 5672 in / 959 out tokens · 19465 ms · 2026-06-26T15:30:35.624852+00:00 · methodology

0 comments
read the original abstract

Vector databases are increasingly used in security sensitive contexts with Retrieval Augmented Generation and organizational AI pipelines; however, their security capabilities remain limited. Specifically, Fine-grained Access Control (FGAC) which is required to ensure that data access adheres to user-specific policies is not fully supported in modern vector databases. Unlike relational databases, vector databases combine structured and unstructured attributes to provide semantic, approximate query results, which complicates FGAC implementation. This creates an inherent tension between enforcing FGAC policies correctly, achieving high ANN search recall and maintaining low query latency. In this paper, we present a vision for Policy-aware Vector Search by formalizing the FGAC policy model in vector databases as well as the enforcement problem. We compare various enforcement strategies, present preliminary findings, and identify key open challenges for future research in policy-aware vector search.

Figures

Figures reproduced from arXiv: 2606.19803 by Lakshmi Sahithi Yalamarthi, Primal Pappachan.

Figure 1
Figure 1. Figure 1: Recall and Latency versus Selectivity for different [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

16 extracted references · 6 canonical work pages

  1. [1]

    d.].arXiv Dataset

    [n. d.].arXiv Dataset. https://www.kaggle.com/datasets/Cornell-University/arxiv

  2. [2]

    Aserto. [n. d.]. ABAC-RBAC.https:// www.aserto.com/ blog/ rbac-vs-abac- authorization-models([n. d.])

  3. [3]

    Yuzheng Cai, Jiayang Shi, Yizhuo Chen, and Weiguo Zheng. 2024. Navigat- ing Labels and Vectors: A Unified Approach to Filtered Approximate Nearest Neighbor Search.Proc. ACM Manag. Data2, 6, Article 246 (Dec. 2024), 27 pages. doi:10.1145/3698822

    work page doi:10.1145/3698822 2024

  4. [4]

    Siddharth Gollapudi, Neel Karia, Varun Sivashankar, Ravishankar Krishnaswamy, Nikit Begwani, Swapnil Raz, Yiyong Lin, Yin Zhang, Neelam Mahapatro, Premku- mar Srinivasan, Amit Singh, and Harsha Vardhan Simhadri. 2023. Filtered- DiskANN: Graph Algorithms for Approximate Nearest Neighbor Search with Filters.10.1145/3543507.3583552(2023), 3406–3416. https://...

    work page doi:10.1145/3543507.3583552(2023 2023

  5. [5]

    Andrew Kane. 2024. pgvector: Open-source vector similarity search for Postgres. https://github.com/pgvector/pgvector

    2024

  6. [6]

    Sean Wang

    Yanjun Lin, Kai Zhang, Zhenying He, Yinan Jing, and X. Sean Wang. 2025. Survey of Filtered Approximate Nearest Neighbor Search over the Vector-Scalar Hybrid Data. arXiv:2505.06501 [cs.DB] https://arxiv.org/abs/2505.06501

    work page arXiv 2025

  7. [7]

    Yu A Malkov and Dmitry A Yashunin. 2018. Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs.IEEE transactions on pattern analysis and machine intelligence42, 4 (2018), 824–836

    2018

  8. [8]

    Primal Pappachan, Roberto Yus, Sharad Mehrotra, and Johann-Christoph Freytag

  9. [9]

    VLDB Endow.13, 11 (2020), 2424–2437

    Sieve: A Middleware Approach to Scalable Access Control for Database Management Systems.Proc. VLDB Endow.13, 11 (2020), 2424–2437. http://www. vldb.org/pvldb/vol13/p2424-pappachan.pdf

    2020

  10. [10]

    Primal Pappachan, Shufan Zhang, Xi He, and Sharad Mehrotra. 2022. Don’t Be a Tattle-Tale: Preventing Leakages through Data Dependencies on Access Control Protected Data.Proc. VLDB Endow.15, 11 (2022), 2437–2449. https: //www.vldb.org/pvldb/vol15/p2437-pappachan.pdf

    2022

  11. [11]

    Liana Patel, Peter Kraft, Carlos Guestrin, and Matei Zaharia. 2024. Acorn: Perfor- mant and predicate-agnostic search over vector embeddings and structured data. Proceedings of the ACM on Management of Data2, 3 (2024), 1–27

    2024

  12. [12]

    2021.Sentence-Transformers: all-MiniLM-L6-v2

    Nils Reimers. 2021.Sentence-Transformers: all-MiniLM-L6-v2. https://huggingface. co/sentence-transformers/all-MiniLM-L6-v2

    2021

  13. [13]

    Qihua Wang, Ting Yu, Ninghui Li, Jorge Lobo, Elisa Bertino, Keith Irwin, and Ji-Won Byun. 2007. On the correctness criteria of fine-grained access control in relational databases(VLDB ’07). VLDB Endowment, 555–566

    2007

  14. [14]

    Naikuo Yang, Howard Barringer, and Ning Zhang. 2007. A Purpose-Based Access Control Model. InThird International Symposium on Information Assurance and Security. 143–148. doi:10.1109/IAS.2007.29

    work page doi:10.1109/ias.2007.29 2007

  15. [15]

    Hongbin Zhong, Matthew Lentz, Nina Narodytska, Adriana Szekeres, and Kexin Rong. 2026. HONEYBEE: Efficient Role-based Access Control for Vector Databases via Dynamic Partitioning. 26 pages. doi:10.1145/3786625

    work page doi:10.1145/3786625 2026

  16. [16]

    Jiaxu Zhu, Jiayu Yuan, Kaiwen Yang, Xiaobao Chen, Shihuan Yu, Hongchang Lv, Yan Li, and Bolong Zheng. 2025. An Experimental Evaluation of Hybrid Querying on Vectors.Proc. VLDB Endow.19, 2 (Oct. 2025), 183–195. doi:10.14778/3773749. 3773757

    work page doi:10.14778/3773749 2025