Statistical Disclosure Control for External Ai via Transformation Groups

Small enterprises increasingly rely on external AI for data analytics, which exposes sensitive summary statistics to the AI provider.We propose a practical framework for statistical disclosure control.It lets enterprises use advanced AI while withholding the statistics they deem confidential.Each data owner first applies a random, reversible transformation drawn from a pre-verified group, and submits only the transformed data.An inverse mapping then recovers the target statistics: the recovered output has the same distribution as the target.We introduce notions of traceability and transitivity for transformation groups and verify them through a two-step procedure.A multi-sample test based on Haar-measure sampling checks traceability; an orbit-based analysis then confirms transitivity.We implement the framework on a pottery-company dataset with five LLMs.In each task the test is not rejected under a suitable model and sampling measure, so that the external AI cannot identify the protected statistics from the submitted data.

Authors

Institutions

Publication Details

Journal
Statistica Sinica
Published
2026-11-17
DOI
https://doi.org/10.5705/ss.202026.0110
Primary Topic
Privacy-Preserving Technologies in Data
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Statistical Disclosure Control for External Ai via Transformation Groups

Lingxuan Shao
1 links
Statistica Sinica
Privacy-Preserving Technologies in Data
article

Statistical Disclosure Control for External Ai via Transformation Groups

Lingxuan Shao
article en

Abstract

Small enterprises increasingly rely on external AI for data analytics, which exposes sensitive summary statistics to the AI provider.We propose a practical framework for statistical disclosure control.It lets enterprises use advanced AI while withholding the statistics they deem confidential.Each data owner first applies a random, reversible transformation drawn from a pre-verified group, and submits only the transformed data.An inverse mapping then recovers the target statistics: the recovered output has the same distribution as the target.We introduce notions of traceability and transitivity for transformation groups and verify them through a two-step procedure.A multi-sample test based on Haar-measure sampling checks traceability; an orbit-based analysis then confirms transitivity.We implement the framework on a pottery-company dataset with five LLMs.In each task the test is not rejected under a suitable model and sampling measure, so that the external AI cannot identify the protected statistics from the submitted data.

Statistica Sinica
Fudan University (CN)
Shanghai Science and Technology Development Foundation, National Natural Science Foundation of China
Privacy-Preserving Technologies in Data
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon