Feedzai Expands Digital Trust to Detect AI Agents in Bank Accounts

Feedzai expanded its Digital Trust product on November 12 so banks can tell whether a person or an AI agent is behind a customer session, extending AI agents bank account fraud detection from account opening to logged-in activity.

The release matters to fraud and financial-crime teams because customers are starting to hand tasks to software that logs in on their behalf. A bank that cannot tell a customer’s own assistant from a criminal’s bot cannot set sensible authentication, consent or liability rules for either.

Feedzai said the update is available now and that it will keep expanding agentic detection in the coming months. The announcement names no bank customers and gives no pricing.

How AI agents bank account fraud detection works

Feedzai said Digital Trust combines the signals an agent declares about itself with device fingerprinting and behavioral signals that point to automated activity. Persistent device and network signals let the system recognize returning activity over time.

The risk score draws on Feedzai’s network view and shared intelligence across its customers. Feedzai also uses cryptographic device binding, which ties each session to a known customer and device. That lets a bank see on whose behalf an agent is acting.

Feedzai markets the product as a single view of trust across cybersecurity, identity and fraud. The same session data serves all three teams, which matters for banks that run them as separate functions.

Telling a customer’s agent from a malicious bot

The hard part of AI agents bank account fraud detection is that the two kinds of agent behave alike. Pedro Barata, Chief Product Officer at Feedzai, put it this way:

“Personal and malicious agents can look very similar.” Pedro Barata, Chief Product Officer, Feedzai

Feedzai’s answer is to recognize an agent once and remember the relationship. Where bank policy allows, a returning agent acting for a known customer on a bound device is not challenged again. Banks keep control over which agent activity is permitted and how customer consent is handled.

A trusted device can also act as a possession factor, which Feedzai said reduces reliance on SMS one-time passcodes. The company said the protection covers login, account recovery, new device enrollment and call center authentication linked to a digital session, including attempts to bypass passkeys.

Feedzai’s performance figures for fraud detection

Feedzai published three results. Using shared signals, it said it detected 44% more fraud in the highest-risk 0.1% of sessions with no increase in false positives.

It also said the product produced 6,000 fewer unnecessary authentication challenges per million logins than competing technology, and 41% fewer SMS one-time passcodes.

These are Feedzai’s own figures. The release does not name the competing technology or identify a customer that produced the results, so banks will want to test them against their own session data.

Who pays when an agent moves the money

No rulebook yet settles liability for a payment an AI agent initiates. Existing frameworks turn on whether the customer authorized a payment and whether the bank had adequate fraud controls.

In the UK, the Payment Systems Regulator’s reimbursement regime for authorised push payment (APP) fraud has applied since October 2024. It caps reimbursement at £85,000 per claim on Faster Payments and splits the cost 50:50 between sending and receiving firms, according to the PSR’s policy statement PS24/7. Teams building their processes can start from RegTech Post’s APP fraud reimbursement template pack.

In the EU, Council and Parliament negotiators reached a provisional agreement on the third Payment Services Directive (PSD3) and the Payment Services Regulation (PSR) on November 27, 2025. Under it, a payment service provider that fails to implement appropriate fraud prevention mechanisms is liable for its customers’ losses. Refunds also apply where a fraudster impersonates the provider’s staff. The agreement still needed technical work before formal adoption.

Authentication adds a second question. Strong customer authentication assumes a person proves who they are with two factors. When a customer delegates a task, the bank must decide whether the agent inherits that authentication or needs its own. Device binding is one answer, because it ties the agent’s session back to a customer-enrolled device.

What supervisors have said on agentic AI

The Financial Conduct Authority published the Mills Review on July 6, 2026. Led by Sheldon Mills, it found that retail finance is moving from human-led activity toward services that are AI-enabled, continuous and delegated.

The FCA’s research found that about a fifth of UK adults, equivalent to 11 million people, are likely to use AI that can act autonomously within pre-set goals. The review named the amplification of fraud and cyber risks as one of four major AI-driven shifts, and recommended that the regulator enable the foundations for agentic finance.

The FCA said it does not plan AI-specific regulation. It relies on its principles-based, outcomes-focused approach, including the Consumer Duty and the Senior Managers Regime. Details are on the FCA’s page for the review.

For senior managers, that means existing accountability applies to agent traffic without new rules telling them how to treat it.

Competition in agentic fraud detection

Feedzai is not alone in AI agents bank account fraud detection. BioCatch publishes an agentic account takeover use case and has launched AiRE for agentic AI fraud detection. Banks are also deploying agents of their own, covered in RegTech Post’s report on banks turning to AI agents for AML and fraud monitoring.

Feedzai’s announcement says the work will continue in the coming months. The next fixed date for buyers is formal adoption of the EU payments package, which will settle the liability wording that banks in the bloc must build their agent policies around.