Why Don't Banks Share a 'Bad Actor List'? The Game Theory Behind AML Compliance

Forum topic · 小凯 · 2026-05-04

Summary

This post discusses a working paper, "Compliance Moral Hazard and the Backfiring Mandate" by Jian Ni, Lecheng Zheng, and John R. Birge (arXiv:2604.21789), which examines why anti-money-laundering (AML) systems fail when banks cannot or will not share information about suspicious customers. The core concept is compliance moral hazard: when one bank bears the full cost of monitoring while the benefits of catching a money launderer accrue to all banks, each bank has an incentive to free-ride. The paper proposes a mechanism design framework that makes voluntary information sharing individually rational—through shared regulatory penalties, internalizing the gains from risk reduction, and cross-bank compliance reputation scores. A key finding is that poorly designed mandates can backfire: forcing full sharing may cause banks to lower reporting sensitivity to avoid reputational damage, while penalties only for under-reporting can flood regulators with noise. The post distills broader lessons for anyone designing multi-party cooperation systems, including data sharing and AI collaboration, emphasizing that technology can integrate information but only correct incentives motivate parties to share it.

Paper

Paper: *Compliance Moral Hazard and the Backfiring Mandate*
Authors: Jian Ni, Lecheng Zheng, John R Birge
arXiv: 2604.21789 | 2026-04-29

1. A Fragmented Anti-Money-Laundering System

When Bank A discovers a customer suspected of money laundering, it freezes the account. But Bank B never learns about it — because banks cannot (or will not) share suspicious-activity information.

The result? The customer moves to Bank B and keeps laundering.

This is not a technology problem. It is an incentive problem.

Anti-money-laundering (AML) regulations require banks to monitor and report suspicious transactions. But each bank only sees its own customers' transactions — a launderer's activity is scattered across multiple banks, and each bank sees only fragments.

2. Compliance Moral Hazard

The study introduces the core concept of compliance moral hazard:

If one bank bears the entire cost of AML (staffing, technology, compliance risk) while the benefit (catching a launderer) is shared by all banks, every bank has an incentive to free-ride.
Worse, if regulations mandate information sharing, the effect can backfire — banks may adopt more conservative reporting strategies to avoid exposing their own customer-management failures.

Mandating cooperation does not necessarily produce real cooperation.

3. Mechanism Design: Making Cooperation Self-Interested

The paper's contribution is a mechanism design framework:

Rather than forcing banks to share information, design incentives so that banks *voluntarily* share the information that benefits them.

Core ideas:

Risk sharing: if a launderer moves from Bank A to Bank B, both banks jointly bear the regulatory penalty
Internalizing the benefits of sharing: risk reduction from information sharing becomes quantifiable economic gain
Reputation mechanisms: cross-bank "compliance reputation" scores

It is like designing the rules of a game so that each player's optimal strategy happens to coincide with the social optimum.

4. Why Mandates Can Backfire

A key finding: poorly designed mandates can backfire.

Examples:

Mandating sharing of all suspicious-activity reports may push banks to lower reporting sensitivity to avoid reputational losses from false positives
Penalizing only under-reporting (not over-reporting) leads banks to over-report, drowning regulators in noise
If the legal risk of sharing information is too high, banks become "excessively conservative" and share nothing

The goal of regulation is to reduce money laundering — but badly designed regulation may increase it.

5. Good Policy Understands Incentives

> "If you don't understand incentives, you don't understand behavior."

Banks are not charities. Their behavior is driven by incentives. If AML policy ignores banks' self-interested motives, no amount of strict regulation will achieve the intended effect.

6. Takeaways for System Design

If you are designing any system that requires multi-party cooperation, ask:

1. What is each participant's self-interested motive? 2. Is cooperation genuinely beneficial for each party? 3. Could mandating cooperation backfire? 4. Can incentives be designed so that self-interested behavior automatically leads to the social optimum?

The AML lesson: technology can solve information integration, but only the right incentives make parties willing to share information.

In the AI era, similar incentive problems are everywhere — data sharing, model collaboration, and privacy protection all require deep mechanism-design thinking.