Fraud prevention startup Darwinium UK Ltd. today launched two new capabilities that make intent intelligence the foundation of its platform.

Journey Transition Probability flags unusual changes in the steps a person, bot or artificial intelligence agent takes on the way to a login, account change or payment. MCP Protection watches the tools agents use to carry out tasks.

The release is aimed at a problem agentic commerce is making harder to manage, since only about one in four agentic transactions on Darwinium’s network self-declares. Purchases involving agents are rejected nine times as often as others.

In a survey Darwinium ran of 500 fraud, risk and security leaders in the U.S. and the U.K., 97% reported an increase in AI-driven attacks. Only 36% believed they had effective fraud coverage across the full customer journey. The company argues that businesses need a way to recognize legitimate agent activity without giving risky actions a free pass.

An authorized AI agent “can start out doing exactly what a customer asked, then take an unexpected turn,” said Michael Rodriguez, chief operating officer at Darwinium. He made the same point about authenticated human customers, who can still be coached into sending money to a scammer. Darwinium has spent years mapping the full customer journey for that reason, Rodriguez said, and making intent intelligence central lets businesses assess an interaction while it is still unfolding.

The company launched its agent intent tools in March, and the existing Agent Intent Detection product already identifies AI agents even when they do not announce themselves. Today’s update folds the tool calls an agent makes over the Model Context Protocol into the same journey as a customer’s web and mobile activity. One model assesses every step. Risk decisions run through the content delivery network the business already uses for its website traffic.

Inside that combined view, Journey Transition Probability measures the order and timing of actions against the customer’s typical behavior. The business’ own traffic and the type of journey underway also factor in. A step that looks ordinary on its own can read as suspicious once the full path is taken into account.

MCP Protection handles the agent side by linking each tool call to the journey that led to it. With it, businesses can verify an agent’s credentials and keep watch on what the agent does once it is working. A payment or other higher-risk action can be held for additional checks.

Jon Ferrari, senior manager of fraud prevention and application security at Darwinium customer Apollo.io, said web traffic has reached “an inflection point where user-agent declarations and even statements of intent are becoming moot.” His team needs to track whether behavior matches stated intent over time, he said. That includes activity in aggregate, where many requests can add up to an outcome no individual request disclosed.

Darwinium is a venture capital-backed startup whose investors include U.S. Venture Partners, Blackbird Ventures Pty. Ltd., Airtree Ventures Pty. Ltd. and Accomplice. The company announced an $18 million round in October 2023.

Image: Darwinium