Abstract
AI regulations are scattered across hundreds of PDFs, guidance notes, and consultations spanning multiple jurisdictions, each with its own structure, terminology, and compliance deadlines. Every compliance AI team and AI governance platform re-solves the same parsing problem independently, extracting obligations from legal texts that were never designed for machine consumption. This duplication of effort is wasteful and error-prone, and it creates an invisible tax on every organisation that needs to verify compliance across more than one regulatory regime. RegGraph addresses this problem by providing shared infrastructure: a queryable, versioned, open dataset of regulatory obligations. It structures AI regulatory requirements into machine-readable JSON, cross-referenced by jurisdiction, regulation, risk category, obligation type, sector, and entity. No PDF parsing is required. The current release covers 45 obligations across 10 regulations and 4 jurisdictions including the European Union (36 obligations from the EU AI Act), the United States (4 obligations spanning HIPAA, SEC AI Guidance, Executive Order 14110, and FTC Act Section 5), the United Kingdom (3 obligations from the UK AI framework, UK GDPR, and the Online Safety Act), and Nigeria (2 obligations from the NDPR and the Nigeria Data Protection Act 2023). Each obligation conforms to a formal JSON Schema that enforces type safety and structural completeness. The dataset is available as both a CLI tool and a JavaScript library under an Apache 2.0 license, enabling programmatic access to regulatory requirements for compliance automation, AI governance tooling, and academic research. Contributions are welcome via pull requests and must include citations to the primary legal source, ensuring the dataset remains authoritative and legally grounded.