Darwinium has launched two capabilities that apply intent intelligence to fraud prevention across human, bot, and AI agent journeys.
US-based Darwinium announced that intent intelligence now forms the basis of its cyberfraud prevention platform. The company links the change to AI agents increasingly carrying out tasks that customers used to perform themselves. Two new features put the approach into practice. Journey Transition Probability identifies irregular changes in the sequence of steps leading to a login, account action, or payment, whether a human, bot, or AI agent is involved. MCP Protection applies the same oversight to the tools AI agents use to complete tasks.
The coverage gap in agentic commerce
Citing data from a report on the rise of AI fraud and agentic commerce, Darwinium shows that 97% of fraud, risk, and security leaders reported a rise in AI-driven attacks. Only 36% considered their fraud coverage effective across the full customer journey. According to the company, this gap becomes harder to manage as agent activity increases.
Data from Darwinium's network indicates that roughly one in four agentic transactions self-declares. Transactions involving agents are rejected nine times as often as other purchases. The company describes the challenge for businesses as recognising legitimate agent activity without exempting higher-risk actions from scrutiny.
Michael Rodriguez, Chief Operating Officer at Darwinium, noted that an authenticated customer can still be coached into transferring funds to a scammer. Likewise, an authorised AI agent may begin by following a customer's instructions and then deviate from them. In the official’s view, placing intent intelligence at the centre of the platform allows businesses to assess an interaction as it unfolds, regardless of whether a person or an agent takes the next step.
How the capabilities work
The launch builds on Darwinium's existing Agent Intent Detection, which identifies AI agents, including those that do not disclose themselves. The company now combines a customer's web and mobile activity with an agent's Model Context Protocol (MCP) tool calls into a single journey. One model assesses each step of that journey, and risk decisions are executed through the business's existing content delivery network (CDN), the infrastructure that already handles its website traffic.
Journey Transition Probability compares the order and timing of actions with three reference points: the customer's typical behaviour, the business's own traffic, and the type of journey underway. The aim is to surface activity that appears routine at a single step but becomes suspicious when viewed across the full path.
MCP Protection links an agent's tool calls to the journey that preceded them. Businesses can check agent credentials, observe the actions an agent performs, and require additional checks before higher-risk actions, such as payments, are completed.
The approach shifts risk assessment from isolated checkpoints to the full sequence of interactions, a change Jon Ferrari, Senior Manager, Fraud Prevention and Application Security at Apollo.io, described as relevant to current traffic patterns. Ferrari said Darwinium has increased the company's visibility into user intent and behaviour across surfaces, for both agents and humans.
