Feedzai adds tools to help banks recognize AI agents accessing customer accounts
Digital Trust links banking sessions to known customers and devices. Banks still decide which agent activity to permit—and when to demand another check.
Loading page…
Digital Trust links banking sessions to known customers and devices. Banks still decide which agent activity to permit—and when to demand another check.
Listen to this story
Banks—not Feedzai’s detection system—still decide whether a recognized AI agent may act, while the expanded Digital Trust platform gives them more ways to link sessions to customers and assess risk. Feedzai reports 44% more fraud detected in the highest-risk 0.1% of sessions without more false positives, alongside 6,000 fewer unnecessary challenges per million logins and 41% fewer SMS codes. The company has not disclosed test conditions or accuracy in distinguishing authorized agents from malicious ones; more agent-detection capabilities are planned in the coming months.
Agent recognition combines declared agent information with device fingerprints and behavioral signals, rather than relying on an agent to identify itself.
Persistent device recognition and network-wide fraud signals help banks assess activity across sessions and spot risks they may not see alone.
Cryptographic device binding links a session to a known customer and device, while banks retain control over whether to allow, challenge, or block activity.
An AI agent can enter a bank account with the right credentials without settling whether its activity should be trusted. Feedzai’s November 12 expansion of Digital Trust aims to help banks make that distinction, combining device recognition, behavioral signals and cryptographic links to customers while leaving banks in control of what agents may do.
The security problem goes beyond stolen passwords. Feedzai says agents can sign in with valid credentials, while criminals can exploit compromised devices, account recovery and social engineering to manipulate real customers. Its response is to assess risk before login and throughout the banking session, rather than treat successful authentication as the end of the check. The system draws on device, behavioral and network information.
Feedzai describes three connected parts of the update. They address different questions: whether activity comes from a returning device, whether it looks risky across the company’s network, and which customer an agent represents. Agent recognition combines information agents declare about themselves with device fingerprints and behavior that indicates automation; it does not rely only on an agent identifying itself.
Recognition does not automatically mean permission. Banks can allow, challenge or block recognized agent activity under their own policies, as Finovate explains. Feedzai says banks can also recognize previously trusted customer-agent relationships and avoid repeated authentication where policy allows. Control over permitted activity and customer consent remains with the bank, rather than passing to the detection tool.
Personal and malicious agents can look very similar. Our latest advancements in Digital Trust help banks tell them apart, blocking unwanted agents and trusting the ones their customers choose to use.
Pedro Barata, Feedzai chief product officer
The release also targets reliance on one-time passcodes sent by text message. Feedzai says a trusted, cryptographically bound device can act as a possession factor: evidence that the customer has the registered device. That is intended to reduce both the cost and the security exposure associated with SMS codes, while removing some extra authentication prompts.
For passkey protection, the company says Digital Trust confirms that a registered device is present and checks it for signs of compromise or social engineering. The authentication coverage extends beyond routine sign-ins to account recovery, new-device enrollment and call-center authentication linked to a digital session. These are safeguards within the same release, not a separate agent-access product.
Feedzai reports that shared signals detected 44% more fraud within the highest-risk 0.1% of sessions, without increasing false positives—legitimate activity incorrectly flagged as fraud. It also reports 6,000 fewer unnecessary authentication challenges per million logins compared with competing technology, alongside 41% fewer SMS one-time passcodes.
Those are company-reported fraud and authentication results, not a published accuracy rate for separating customer-authorized agents from malicious ones. The release does not name the competing technology or give a test period or sample size. That limits how closely a bank can compare the headline gains with its own environment.
Feedzai describes this expansion as a foundation for detecting and managing agents at scale, with more agent-detection capabilities planned over the coming months. For banks, the immediate offering combines recognition and authentication controls; permission to act on an account still follows each bank’s rules.
Loading discussion...
Join the conversation
Explain where customer choice should give way to bank security rules.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.