Never Let Your Agent Touch the Database

Version 1.1 is a vocabulary revision, content-preserving: aligned with The Theory of Data, Version 4.0 — a measure is the governed family; a member of a measure is a governed series; a series is a typed binding of datums over one anchor. Nothing is altered in substance.AI agents can be genuinely useful in analytical systems — interpreting questions, searching governed models, proposing structured requests, explaining results. They should not hold database credentials or control an executable database language. This position argues the prevailing text-to-SQL architecture concentrates three risks in one authority (security, cost, and — least discussed — analytical correctness: queries that execute perfectly and are silently wrong), and that guardrails inspect commands while the failures live in meanings. The alternative is an intent boundary: the model proposes a declaration of the desired result; a governed environment resolves identities, certifies obligations, and owns execution. The model interprets; it never authors the final command. Fifth in the datumwise position series.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-08-03
DOI
https://doi.org/10.5281/zenodo.21774490
Primary Topic
Advanced Database Systems and Queries
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
preprint

Never Let Your Agent Touch the Database

Huayin Wang
Zenodo (CERN European Organization for Nuclear Research)
Advanced Database Systems and Queries
preprint

Never Let Your Agent Touch the Database

Huayin Wang
preprint en

Abstract

Version 1.1 is a vocabulary revision, content-preserving: aligned with The Theory of Data, Version 4.0 — a measure is the governed family; a member of a measure is a governed series; a series is a typed binding of datums over one anchor. Nothing is altered in substance.AI agents can be genuinely useful in analytical systems — interpreting questions, searching governed models, proposing structured requests, explaining results. They should not hold database credentials or control an executable database language. This position argues the prevailing text-to-SQL architecture concentrates three risks in one authority (security, cost, and — least discussed — analytical correctness: queries that execute perfectly and are silently wrong), and that guardrails inspect commands while the failures live in meanings. The alternative is an intent boundary: the model proposes a declaration of the desired result; a governed environment resolves identities, certifies obligations, and owns execution. The model interprets; it never authors the final command. Fifth in the datumwise position series.

Zenodo (CERN European Organization for Nuclear Research)
Peace, Justice and strong institutions
Advanced Database Systems and Queries
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.