Banks Are Using AI on Fraud Cases, but Struggle to Prove How It Decides

Banks are using artificial intelligence to help handle fraud cases and customer disputes. The harder part is proving how it influenced a decision. New state laws in Texas and Colorado raise the bar on records and oversight, and experts at a recent PaymentsJournal webinar said many banks are not ready.

What AI is doing now

Banks face a flood of alerts, disputes and paperwork. AI helps by summarizing cases and spotting odd patterns in large amounts of data. David Chmielewski of Quavo said it now reads evidence-heavy dispute documents and sorts incoming customer messages. It also helps with investigations, but always with a person involved. Handing everything to AI is, in his words, "not quite yet realized."

What the new rules ask for

Texas's Responsible Artificial Intelligence Governance Act took effect this year. It bans harmful and intentionally unlawful uses of AI and encourages established risk management practices. It also creates a sandbox, a space where companies can test AI under state oversight.

Colorado passed the nation's first broad AI law, then replaced it with the Automated Decision-Making Technology Act. That law is scheduled to take effect next year and is described as more practical.

Where banks fall short

Hailey Windham of Sardine.ai said many banks treat AI as a productivity tool, while regulators focus on governance. If a tool affects whether a dispute is denied, an account is frozen or funds are delayed, regulators may count it as part of the decision.

The gap is paperwork. Teams may know a tool works but cannot always explain these things:

• What data the tool uses
• Why it recommended an action
• How often humans override it
• Whether outcomes are monitored across customer groups

AI also changes over time, a problem called model drift. Customer behavior and fraud tactics shift, so Windham said oversight must be continuous, not a one-time exercise. The duty also reaches outside vendors, so banks cannot lean only on risk scores.

Jennifer Pitt of Javelin Strategy & Research added that many mid-sized and smaller banks keep fraud, anti-money-laundering, identity and data teams in separate systems. That raises a basic question of who is responsible for a decision.

Why customers should care

Trust is part of this. Javelin surveys about 5,000 US adults each year on fraud and scams. Pitt said people who are told what AI is used for, and why, are more willing to accept it. The material does not give the full survey figures.

What happens next

Colorado's law is due next year, so banks have limited time to build records. The material does not say how either state will enforce its law or what penalties apply.