Darwinium Unveils New Fraud Prevention Tools Based on Intent Intelligence
An innovative approach to fraud prevention has emerged from startup Darwinium UK Ltd., which today announced the launch of two new capabilities designed to integrate intent intelligence into its platform. These features, dubbed Journey Transition Probability and MCP Protection, aim to enhance the security of transactions conducted by artificial intelligence agents.
The Journey Transition Probability feature identifies unusual patterns in the paths taken by individuals, bots, or AI agents during logins, account modifications, or payment processes. In contrast, MCP Protection scrutinizes the tools utilized by these agents as they execute tasks within various systems. This dual approach addresses the growing complexity and risk associated with agentic commerce, where roughly only 25% of transactions involving agents self-declare. Alarmingly, transactions employing agents are rejected at a rate nine times higher than their human counterparts.
A recent survey conducted by Darwinium involving 500 fraud, risk, and security professionals across the United States and the United Kingdom revealed that 97% observed an increase in AI-driven attacks. Despite this alarming trend, only 36% of respondents felt they had comprehensive fraud coverage throughout the customer journey. Darwinium emphasizes the necessity for businesses to effectively identify legitimate agent activities while simultaneously mitigating the risks posed by suspicious actions.
Michael Rodriguez, Chief Operating Officer at Darwinium, emphasized that even authorized AI agents could initially follow customer instructions before deviating into risky behaviors. This concern extends to authenticated human customers, who may be manipulated into transferring funds to malicious actors. To address these vulnerabilities, Rodriguez stated that Darwinium has invested considerable time in mapping the complete customer journey, allowing for real-time assessment of interactions as they develop.
Having initially unveiled its agent intent tools in March, Darwinium’s updated Agent Intent Detection product now incorporates interactions made by agents over the Model Context Protocol alongside the user’s traditional web and mobile activity. This integrated model evaluates each action within the context of the entire user journey, enabling informed risk assessments guided by existing content delivery networks that support web traffic.
Within this comprehensive framework, Journey Transition Probability analyzes the timing and sequence of actions against a customer’s typical behavior patterns, incorporating the overall traffic data and the context of the specific transaction. An action perceived as benign in isolation may raise red flags when evaluated in the context of the entire path taken by the user.
The MCP Protection feature serves to monitor the agent’s behavior by linking tool calls to their corresponding transactional journey, allowing businesses to authenticate an agent’s credentials and supervise their actions post-authentication. This functionality enables organizations to hold high-risk actions, such as payments, for further scrutiny before approval.
Jon Ferrari, Senior Manager of Fraud Prevention and Application Security at Darwinium client Apollo.io, acknowledged the shifting landscape of web traffic, stating that traditional user-agent declarations and express intents have become increasingly irrelevant. His team is focused on aligning actual user behavior with articulated intent over time, considering aggregate activity where numerous minor requests may contribute to significant outcomes that individual requests do not reveal.
Darwinium, backed by venture capital from investors including U.S. Venture Partners, Blackbird Ventures Pty. Ltd., Airtree Ventures Pty. Ltd., and Accomplice, recently secured $18 million in funding as of October 2023. This funding will enable the ongoing development and refinement of its security technologies designed to combat the rising tide of cyber threats targeting agentic commerce.