Published in the Fordham Law Review 94(1), 57–129, by Cullen O'Keefe, Ketan Ramakrishnan, Janna Tay, and Christoph Winter.
The definition
The article defines AI agents as "AI systems that can perform computer-based tasks as competently as human experts" — a capability-parity definition rather than an architectural one, which sets the scope of the argument at the point where an agent could substitute for an expert.
The argument
The premise is behavioural symmetry: "Humans use computers to commit crimes, torts, and other violations of the law. As AI agents progress, therefore, they will be increasingly capable of performing actions that would be illegal if performed by humans."
The stakes: "Such lawless AI agents could pose a severe risk to human life, liberty, and the rule of law." The third item is the distinctive one — the concern is not only that agents cause harm but that they erode the mechanism by which harm is ordinarily constrained.
The proposal, which the authors describe as "a simple claim": "in high-stakes deployment settings, such as government, AI agents should be designed to rigorously comply with a broad set of legal requirements."
Two features bound it. It is scoped to high-stakes deployment settings, government named first, rather than proposed as a universal design constraint. And it targets design rather than liability — building compliance into the agent rather than assigning responsibility after a violation, which distinguishes it from the non-delegable-liability approach in SCSP's agentic-governance assessment.
The framing of the field: "designing public policy for AI agents is one of society's most important tasks."
Relationships
- related: Agentic AI: The Governance Race We Must Not Lose (SCSP, June 2026) — an alternative approach to the same problem, assigning non-delegable liability rather than designing for compliance
- related: AI Liability, Agentic AI, AI and Tort Liability, Frontier AI Governance
- related: Federal AI Adoption — Patterns and Tensions