Beyond Static Rules - Automated Discovery of Latent Vulnerabilities in Text-to-SQL

Sep 21, 2026

Uploaded by Asher

Large Language Models (LLMs) have shown strong performance on Text-to-SQL, but real-world deployment is constrained by latent reliability weaknesses that existing diagnostics cannot systematically uncover. The approach addresses this gap by replacing static, expert-defined rules with an autonomous guided exploration loop. It generates vulnerability hypotheses, uses an evolving Vulnerability Codex to craft targeted perturbations, and iteratively documents defects. Experiments on open-source models reveal many failure cases, strong cross-model transferability, and preliminary remediation via lightweight fine-tuning.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages17

Reading Time43 min

AccessFree to download

Abstract

1 Introduction

2 Related Work

2.1 LLM-based Text-to-SQL Generation

Beyond Static Rules: Automated Discovery of Latent Vulnerabilities in Text-to-SQL Hanqing Wang 1∗, Yongdong Chi 1 * , Jian Yang2 , Lei Yang3 , Jiehui Zhao3 , Yun Chen 1 , Guanhua Chen4† 1 Shanghai University of Finance and Economics, 2Beihang University, 3Deepexi Technology Co. Ltd., 4 Southern University of Science and Technology arXiv :2607 .03833v 1 [ cs .CL] 4 Jul 2026 Abstract While Large Language Models (LLMs) have achieved remarkable success in Text-to-SQL tasks, their deployment in real-world environments is hindered by latent reliability issues. Identifying these latent weaknesses is critical for building trustworthy database interfaces, yet current diagnostic approaches rely heavily on static, expert-defined rules, which lack the capability for systematic and automated exploration. To bridge this gap, we propose SAGE (Systematic Automated Guided Exploration), a novel framework designed to autonomously uncover latent failure patterns in LLM-based Text-to-SQL generation. Specifically, SAGE generates vulnerability hypotheses for given samples and references a continuously evolving Vulnerability Codex to design targeted perturbations, thereby iteratively verifying and documenting potential defects. Extensive experiments on state-of-the-art open-source LLMs demonstrate that SAGE uncovers a substantial number of failure cases, highlighting the significant fragility of current models. Furthermore, our analysis reveals that the Vulnerability Codex exhibits strong cross-model transferability, indicating that the discovered patterns represent generalized structural weaknesses. Finally, we explore SAGE’s potential for remediation. Although preliminary, lightweight fine-tuning on the generated samples yields promising improvements, suggesting a scalable pathway for closing the reliability loop in future work. 1 Introduction Text-to-SQL systems (Androutsopoulos et al., 1995 ; Li and Jagadish, 2014 ; Li et al., 2024b ; Yu et al., 2018) serve as a vital bridge between natural language and structured data, democratizing data access by enabling non-technical users to query complex databases intuitively. With the recent integration of Large Language Models (LLMs) (Pour- * Equal Contribution.† Corresponding Author. reza and Rafiei, 2023 ; Xie et al., 2025 ; Li et al., 2023a), this field has witnessed a paradigm shift in semantic understanding and execution accuracy. Systematically identifying latent failure modes is the prerequisite for enhancing this reliability. However, current evaluation paradigms predominantly rely on static benchmarks (e.g., Spider (Yu et al., 2018), BIRD (Li et al., 2023b)) or manual error analysis (Chang et al., 2023a ; Gan et al., 2021b,a; Patil et al., 2023 ; Pi et al., 2022) . These approaches are inherently constrained by their reliance on human priors: they primarily verify model robustness against predefined challenges (Zhang et al., 2024 ; Deng et al., 2021 ; Bhaskar et al., 2023), but cannot provide a systematic exploration of potential or previously unidentified vulnerabilities 1. As a result, certain failure modes (Saparina and Lapata, 2024 ; Sahitaj et al., 2025 ; Chang et al., 2023b) may remain insufficiently examined and can have a substantial impact on the reliability of Text-to-SQL generation beyond what is captured by fixed heuristics. Consequently, there is a pressing need for an automated framework capable of proactively exploring the model’s failure boundaries. To bridge this gap, we propose SAGE (Systematic Automated Guided Exploration), a framework designed to proactively expose latent vulnerabilities in LLM-based Text-to-SQL generation. Unlike passive verification methods (Changet al., 2023a ; Gan et al., 2021b,a; Patil et al., 2023 ; Pi et al., 2022), SAGE orchestrates an active discovery process. Specifically, it first formulates specific vulnerability hypotheses for a target sample, and then references a continuously evolving Vulnerability Codex, a dynami

FAQs

What problem does the paper address in LLM-based Text-to-SQL systems?

It targets latent reliability issues that are not fully revealed by static benchmarks or manual error analysis, leaving potential failure modes insufficiently examined.

How does SAGE discover latent vulnerabilities?

SAGE formulates vulnerability hypotheses for given samples and consults a continuously evolving Vulnerability Codex to generate targeted perturbations, then iteratively verifies and refines failure-boundary understanding.

What evidence is provided to show SAGE’s effectiveness and generality?

Experiments on state-of-the-art open-source LLMs show SAGE uncovers substantial failure cases, with analysis indicating strong cross-model transferability of the discovered vulnerability patterns.

Sep 21, 2026

Uploaded by Asher