# Darwinium Puts AI-Agent Intent Analysis at Core of Fraud Detection

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/28082
Published: 2026-10-08T13:08:59.000Z
Updated: 2026-10-08T13:08:59.000Z
Section: Technology

> Its platform evaluates customer journeys and machine tool calls to distinguish legitimate automation from suspicious activity.

Darwinium is making intent analysis a central part of fraud detection for human users, bots and AI agents as automated activity becomes harder to distinguish from legitimate customer behavior.

Its Journey Transition Probability capability models normal customer journeys and evaluates individual actions by their sequence, timing and broader journey context. An action that appears routine by itself can receive greater scrutiny when it breaks from the expected path.

The platform’s Model Context Protocol controls monitor tools called by AI agents, authenticate non-human identities and assess whether those calls match legitimate intent. The Model Context Protocol is an open JSON-RPC standard that lets AI agents discover and call tools on web services.

The platform is designed to operate at the edge across content-delivery networks including Cloudflare and AWS CloudFront. Announced March 10, it distinguishes verified AI agents, human users and malicious automation before permitting, verifying, challenging or preventing activity based on risk.

“The real challenge is distinguishing trusted automation from abuse and applying the right level of trust at the right moment,” Darwinium CEO Alisdair Faulkner said. “Our intent intelligence gives organizations the intelligence to do exactly that.”

A survey of 500 fraud, risk and security leaders in the United States and United Kingdom found that 97% had seen AI-facilitated attacks increase during the previous 12 months. Only 36% believed their organizations had effective end-to-end fraud coverage, while 64% could stop fraud only at some checkpoints or one main checkpoint.

Among the surveyed organizations, 48% allowed agentic traffic by default with monitoring, 31% blocked it by default and 20% handled it case by case. Average annual direct losses from AI fraud were $4.5 million, while false positives produced average annual revenue losses of $3 million.

The approach treats fraud detection as a continuous assessment of a customer journey rather than a series of isolated checks. MCP controls connect an agent’s tool calls with preceding customer activity, verify agent credentials and apply additional checks to higher-risk actions.