← Back
Evidence source 5349Spot Checked

Governing AI Agents: Risk, Compliance, and Accountability in Law and Finance

papers.ssrn.com2025-12-15Book Chapter
Executive summary

A new book chapter presents a framework for governing advanced AI agentic systems, which set goals and adapt independently. The authors argue that traditional software rules fail in high-risk fields like law and finance, where professional duties cannot be handed to AI. Their GPA+IAT model links system properties to governance needs. Notably, AI can discriminate during processes, not just outcomes, and liability usually falls on the deployer. The chapter uses real cases but lacks detailed guidance for small organizations.

What it examines

This chapter explains how to govern AI systems that act independently in law and finance. It shows how to match governance controls to system risks, focusing on accountability, compliance, and professional duties. The goal is to help organizations deploy AI responsibly in regulated fields.

What it concludes

Strong governance is essential for safe, trustworthy AI in law, finance, and auditing. The chapter offers practical steps for risk assessment, monitoring, and accountability. Applications include legal research, credit decisions, and audits. Future work should refine controls as AI evolves and regulations change.

Extracted from this source

Evidence objects

Evidence 472372% extraction confidence
A new framework for governing AI 'agentic systems'which set goals, adapt, and decide independentlyshows traditional software rules fall short, especially in high-stakes sectors like law, finance, and auditing.

key_findings bullet 1 · key_findings · validation V0

Evidence 472472% extraction confidence
The GPA+IAT model maps agentic system propertiesGoal, Perception, Action, Iteration, Adaptation, Terminationto governance needs, while a five-layer 'governance stack' reveals no single rulebook can ensure safe AI deployment.

key_findings bullet 2 · key_findings · validation V0

Evidence 472572% extraction confidence
Surprisingly, AI can discriminate during processes, not just outcomes, and liability almost always lands on the deployer, not the vendor; the chapter urges layered controls but lacks step-by-step guidance for smaller organizations.

key_findings bullet 3 · key_findings · validation V0

Evidence 472672% extraction confidence
This paper synthesizes governance, risk, compliance, and accountability frameworks for AI agents in law and finance, offering a novel perspective by contextualizing established principles within AI deployment. While not introducing new quantitative models or technical innovations, its originality lies in bridging governance concepts with AI, making it relevant for organizational oversight discussions.

key_findings bullet 4 · key_findings · validation V0

Raw abstract and provenance

Agentic AI systems—capable of autonomous goal pursuit, environmental perception, and iterative action—are rapidly entering legal and financial services,

Source row: 998 · abstract type: snippet