Legal infrastructure for transformative AI governance
Clinical Snapshot
PICO Framework
| P — Population | Regulatory bodies, policymakers, legal scholars, and jurisdictions seeking governance frameworks for advanced artificial intelligence systems |
| I — Intervention | Establishment of legal and regulatory infrastructure for AI governance, including frontier model registration regimes, autonomous agent identification regimes, and regulatory markets for private AI regulatory services |
| C — Comparator | Current substantive-rule-focused AI governance approaches without dedicated legal infrastructure |
| O — Outcomes | Adequacy, feasibility, and design principles of legal frameworks capable of governing transformative AI systems |
Bottom Line
This PNAS Perspective by Hadfield makes a conceptually important contribution to AI governance discourse by distinguishing between substantive AI rules (what AI should or should not do) and the legal infrastructure needed to generate, implement, and adapt those rules over time. The author proposes three governance mechanisms — frontier model registration, autonomous agent identification, and private regulatory markets — as foundational infrastructure for transformative AI governance. For healthcare professionals and health system leaders, the paper's core insight is practically significant: without robust regulatory infrastructure, even well-intentioned AI rules will be unenforceable, unadaptable, and ultimately ineffective. The proposals have direct analogues in healthcare AI governance, particularly regarding clinical AI system registration and the accountability of autonomous clinical agents. However, as a non-empirical expert Perspective, the paper carries the lowest level of evidence under CEBM criteria. Its proposals are untested, the cost-benefit analysis is absent, and the US-centric framing requires substantial adaptation for Australian and other international contexts. Clinicians and health policymakers should treat this as a thought-leadership contribution warranting serious engagement, but should not adopt its frameworks without empirical evaluation and jurisdictional adaptation. The paper is best read alongside emerging TGA SaMD guidance and the Australian AI Safety Standard.
Key Findings
Effect Size: Not applicable — no empirical effect size reported
Primary Outcome: The paper argues that effective AI governance requires dedicated legal and regulatory infrastructure — not merely substantive rules — and proposes three infrastructure mechanisms: (1) registration regimes for frontier AI models, (2) registration and identification regimes for autonomous agents, and (3) regulatory markets enabling private companies to deliver AI regulatory services under government oversight
Nnt Or Sensitivity: Not applicable — normative legal analysis; no diagnostic, therapeutic, or prognostic metrics reported
Confidence Interval: Not applicable — no quantitative analysis performed
Clinical Application
The conceptual frameworks proposed are intellectually coherent but operationally underdeveloped for immediate healthcare implementation. Frontier model registration and autonomous agent identification regimes have direct analogues in healthcare AI (e.g., clinical decision support software, autonomous diagnostic systems), but translating these frameworks into workable regulatory instruments requires substantial further development, stakeholder consultation, and legislative action. Australia's AI governance landscape is directly relevant to this paper's proposals. The Therapeutic Goods Administration (TGA) has developed guidance on Software as a Medical Device (SaMD) and AI/ML-enabled medical devices, which partially addresses frontier model registration for clinical AI. The Australian Government's voluntary AI Safety Standard and the Department of Industry's AI governance frameworks represent nascent infrastructure analogous to what Hadfield proposes. The RACGP has published position statements on AI in general practice emphasising transparency and accountability. Autonomous agent identification — particularly relevant to AI scribes, diagnostic assistants, and care coordination tools entering Australian primary and secondary care — remains largely unaddressed in current TGA or AHPRA frameworks. Regulatory market concepts would require careful adaptation given Australia's mixed public-private regulatory architecture and the dominance of Commonwealth-level health regulation. Healthcare policymakers, health technology regulators, hospital executives, clinical informaticists, and health law practitioners engaged in AI governance within healthcare systems
Abstract
Most of our AI governance efforts focus on substance: What rules do we want in place? What limits or checks do we want to impose on AI development and deployment? But a key role for law is not only to establish substantive rules but also to establish legal and regulatory infrastructure to generate and implement rules. The transformative nature of AI calls especially for attention to building legal and regulatory frameworks. In this Perspective, I review three examples: the creation of registration regimes for frontier models; the creation of registration and identification regimes for autonomous agents; and the design of regulatory markets to facilitate a role for private companies to innovate and deliver AI regulatory services.
References
- 1.Hadfield, G. K. (2026). Legal infrastructure for transformative AI governance. Proceedings of the National Academy of Sciences of the United States of America. Advance online publication. https://doi.org/10.1073/pnas.2509742123
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service