Feedzai has expanded its Digital Trust product so banks can identify when AI agents are entering customer accounts, widening a system that now watches for risk before login and throughout the customer journey. The update leans on persistent device recognition, cryptographic device binding and a fraud score built from Feedzai’s network-wide view of activity.
The timing matters because banks are being pushed to tell apart a real customer from an automated agent that may be using the right credentials. In that environment, a valid login no longer means the session is trustworthy. Device compromise, account recovery routes and social engineering can still open the door, which is why the search for stronger trust controls has become more urgent now.
Pedro Barata said the company’s approach starts with protected signals, cryptographic device recognition and network intelligence, because personal and malicious agents can look very similar. He said the latest changes help banks separate them, block unwanted agents and keep the ones customers choose to use. That is the core shift in the product: not just checking whether the password is correct, but deciding whether the session itself deserves trust.
Feedzai says its shared signals found 44% more fraud inside the riskiest 0.1% of sessions without increasing false positives. It also reported 6,000 fewer unnecessary authentication challenges per million logins compared with rival technology, along with a 41% reduction in SMS OTPs. Those numbers point to a trade-off banks have been trying to solve for years: tighter security without turning every interaction into a friction point for customers.
Digital Trust covers login, account recovery, enrolment of new devices and call centre authentication tied to a digital session, giving banks a single view across moments where fraud often slips through. The cryptographic binding ties the customer, device and bank together, which helps make the session harder to hijack even when credentials are genuine. That is also why the update matters beyond the login screen; the weak point is often what happens after the first successful sign-in.
Feedzai plans to build out the agentic detection features over the coming months, but it has not said which banks will adopt them first or when the fuller rollout will land. For banks, the practical question is whether those shared signals can be folded into existing authentication flows quickly enough to replace some SMS OTPs and cut step-up checks without leaving gaps in recovery and support channels. The next test is whether the new trust model can keep up with the speed of the agents it is trying to catch.
