Darwinium launches two intent intelligence capabilities to catch fraud by AI agents

Fraud prevention firm Darwinium UK Ltd. has unveiled two new features that use intent intelligence to bolster its platform against fraud. Journey Transition Probability detects unusual shifts in the sequence of actions taken by users, bots, or AI agents as they progress towards activities like logins, account modifications, or payments.

Meanwhile, MCP Protection keeps tabs on the tools agents employ to complete tasks. The launch aims to tackle a challenge posed by agentic commerce, where only about one in four transactions on Darwinium's network explicitly declare the presence of an agent. These transactions are nine times more likely to be rejected compared to those conducted directly by humans.

A survey of 500 fraud, risk, and security professionals in the U.S. and U.K. revealed that 97% reported a rise in AI-driven attacks, with only 36% feeling confident in their ability to provide comprehensive fraud protection across the entire customer journey. Darwinium argues that businesses require a mechanism to distinguish between legitimate agent behavior and risky actions.

Michael Rodriguez, Darwinium's chief operating officer, explained that an authorized AI agent may initially act in accordance with a customer's request before deviating unexpectedly. He noted a similar scenario for authenticated human customers who could potentially be swayed into transferring funds to scammers. The company has been mapping the entire customer journey for years, and centralizing intent intelligence allows businesses to evaluate interactions in real-time.

Darwinium introduced its agent intent tools in March, and the existing Agent Intent Detection product already identifies AI agents even when they do not disclose their identity. The latest update integrates the tool's calls, which utilize the Model Context Protocol, into the same journey as a customer's web and mobile activity. A single model assesses every step of the journey, considering both the business's historical data and the current nature of the interaction.

Risk decisions are made using the content delivery network that the business already employs for its website traffic. By analyzing the full path of actions, Journey Transition Probability can flag actions that appear ordinary on their own but become suspicious when viewed within the broader context. MCP Protection focuses on the agent side of things, linking each tool call to the journey it belongs to.

This enables businesses to verify an agent's credentials and monitor its subsequent actions. If a payment or other high-risk action is detected, it can be held for further scrutiny. Jon Ferrari, a senior manager at fraud prevention and application security at Darwinium's customer Apollo.io, highlighted that web traffic has reached a critical inflection point where user-agent declarations and even explicit statements of intent are becoming less reliable.

Ferrari's team now needs to track whether behavior aligns with the stated intent over time, even in aggregate scenarios where multiple requests collectively result in an outcome that may not be disclosed by any individual request. Darwinium, a venture capital-backed startup with investors including U.S. Venture Partners, Blackbird Ventures Pty.

Ltd., Airtree Ventures Pty. Ltd., and Accomplice, announced an $18 million funding round in October 2023.