Agent Trust Is a Three-Legged Race: A Conversation on What to Verify Before
AI agents are already in digital banking. They are starting to open accounts, move money, and manage logins. Last year, automated traffic grew about 8 times faster than human traffic.
Fast, exact, and nonstop used to be the fingerprint of fraud. Now it’s the fingerprint of a customer’s helpful assistant. And the signals you leaned on to tell them apart are the easiest ones for an agent to cop
So what can you trust? Not the agent alone. Not the person alone. The tie between them.
We call it binding: proof that a real person put this agent to work. Without it, you can’t trace what an agent did or who answers for it. A real customer’s login can sit inside a hijacked tool. A stolen identity can drive a legit assistant. Check only one side, and both look fine.
Chris Nygard (KeyBank), Kumar Senthil (Firmly.ai), and Nate Soffio (Prove) discuss securing the autonomous bank. Three seats, three legs of the race: the bank, the agent platform, and the identity layer. One shared question: How do we protect customers and help them make the most of AI? Hosted by AboutFraud on November 17.
What you’ll hear
- Binding 101. Why checking the agent or the person alone leaves a gap.
- Good bot, bad bot. Four cases, from “serve it” to “fraud with better tools.” Know what to allow, challenge, or block.
- Both sides of the handoff. How an agent platform proves a person is behind an order, and what a bank needs to see before it says yes.
- The permission slip. What a strong one includes: scoped, expiring, and easy to revoke.
- Who answers for the agent? How a proven tie helps in a dispute or chargeback.
Chris Nygard
Kumar Senthil
Nate Soffio
