The paper

Title: Superviz26-SQL: A Multi-Domain Benchmark for SQL Attack Detection
Authors: Grégor Quetel, Pierre-François Gimenez, Thomas Robert, Laurent Pautet
Venue: Assessment with New methodologies, Unified Benchmarks, and environments, of Intrusion detection and response Systems (ANUBIS) 2026. Abstract: Machine-learning intrusion detectors that excel on a testbed often fail once the deployment environment changes. Yet suitable datasets that comprise several comparable domains under controlled distribution shift remain scarce. This observation holds for the emblematic SQL attack detection (SQLAD), where public clear-text data is confined to single application domains, leaving the cross-domain behaviour of detection methods unmeasured. We introduce Superviz26-SQL, the first multi-domain SQLAD benchmark, built by extending the synthetic generation methodology of Superviz25-SQL to four heterogeneous domains derived from real-world database projects. From the four domains we derive a ready-to-use dataset for each of three evaluation protocols: Superviz26-SQL-LODO for cross-domain generalisation, Superviz26-SQL-CD for same-domain concept drift, and Superviz26-SQL-FSL for few-shot adaptation. We confirm through a lexical-field analysis that the domains exhibit genuine inter-domain diversity, and that each training set is large enough to train deep detectors. We illustrate each protocol using reference baselines spanning five feature extractors and three decision engines. Superviz26-SQL, its three derivatives, and all generation and experiment code are publicly released to support reproducible cross-domain SQLAD research.
Pre-print: TBD
The dataset: Zenodo link.