Agentic AI in Law and Finance

A paper from Bommarito, Katz, and Bommarito asks how to govern AI agents once they start acting on their own inside law and finance. The authors have spent years on how AI actually shows up in legal and financial work, and this is their framework for keeping autonomous systems in bounds.

Their answer is a five-layer regulatory stack. Foundational law sits at the base, then professional ethics, then sector-specific regulation, then the newer AI-specific rules like the EU AI Act, and finally voluntary assurance standards on top. Oversight scales to the risk profile of the task. The paper works through the design questions that come with running agents in high-stakes settings: what the system takes as input, how it executes, and what keeps it safe.

What I take from it is that governance for agents has to work as a stack. An agent can be fully capable and still owe answers to several layers at once, from black-letter law to whatever assurance the profession agrees to. For legal and finance teams, the work ahead is figuring out which layer a given decision answers to, and being able to show it.