Skip to content

USE CASE

Agentic AI Risk

Know your agent: Good AI or Bad AI, before the transaction.


AI agents now log in, navigate banking apps and initiate payments, on behalf of customers and on behalf of criminals. The session may look authenticated.

The question is who, or what, is really driving it, and why.

4.5x
AI-enhanced fraud is 4.5 times more profitable for criminals.

– Interpol

22  agentic fraud use cases mapped

5  critical agentic fraud themes

Agentic fraud is the newest and fastest-moving fraud category. AI makes mediocre scammers adequate: agents automate account takeover, operate the victim's own device, coach victims into payments and blend human and machine within one attack. Agentic AI Risk reads every one of those sessions from login to payment.

ThreatFabric has mapped Agentic AI Risk against 22 agentic fraud use cases.

The hidden signals behind agentic activity

An agent can claim to be anything. The device it runs on, the session it creates and the way it behaves reveal what it really is.

Automation in the session

Headless browsers, injected input and interaction that is too regular or too fast to be human.

Behaviour unlike the customer

Sessions that deviate from how this customer, or this customer's own agent, normally banks.

Apps controlling the device

Accessibility services, agent apps and remote access tools operating the banking app.

Intent across the journey

Login, navigation and payment read as one journey, not as isolated checks.

How ThreatFabric detects agentic fraud

Agentic AI Risk in the Fraud Risk Suite (FRS) combines four propositions: Device Risk, Behavioural Analytics, the new Agentic AI Model and Mobile Threat Intelligence. Together they tell the customer's own agent, doing what it was asked, apart from one that is hijacked, run by a fraudster, or coaching a victim. That's Know Your Agent (KYA).

DEVICE RISK

What is driving the device?

Detects accessibility-driven automation, malware, remote access tools, emulators and device farms, and correlates them with behaviour in the session.

BEHAVIOURAL ANALYTICS

Human or agent? Customer or fraudster?

The patented Identity Model learns how each customer, and each customer's own agent, normally drives a session. The Fraudster Model recognises patterns from confirmed fraud. Together they tell who is in control.

AGENTIC AI MODEL

Agent impersonation and agent intent?

Recognises agents posing as humans or as the customer's own agent, and reads whether a legitimate agent is doing what it was asked, or being turned into a good agent doing bad things.

MOBILE THREAT INTELLIGENCE

See new agentic tooling first

Tracks the malware families and agent apps abusing banking apps, such as MCP-based Android apps automating transfers, and feeds them into detection.

Every path an agent takes into the app

As Android opens up to AI agents, Agentic AI Risk works regardless of how the agent reaches the banking app.

AppFunctions

The bank publishes the actions an agent may call.

Computer Control

A built-in assistant operates the app from screenshots, with no bank integration.

Accessibility abuse

The path ThreatFabric already sees exploited on real devices today.

Why ThreatFabric is effective against agentic fraud

Focus on the complete session journey

  • Tells humans, good agents and bad agents apart

  • Reads agentic intent from login through to payment

  • Detects agents disguising themselves as humans

  • Correlates agent behaviour with malware and device risk

  • Spots coaching and hybrid agent-human attacks

  • Lets banks say yes to the customer's own agent, instead of blocking every agent

  • Protects mobile and web channels

Traditional controls ask:
"Is this an agent?"

ThreatFabric asks:
"Is this the customer's own agent, doing what it was asked?"

Fraud Prevention Built On
Actionable Threat Intelligence

Learn how Agentic AI Risk and Know Your Agent can protect your customers and your organisation.