Darwinium, an AI fraud prevention platform based in the US, has unveiled Agent Intent Intelligence. This solution is tailored to manage security risks related to agentic commerce, where AI agents perform actions like browsing, purchasing, and account management on behalf of consumers.

Deployed at the edge through major CDN providers such as Cloudflare and AWS CloudFront, Darwinium’s Agent Intent Intelligence evaluates each request in real time. It discerns whether a request is from a verified AI agent, a human user, or malicious automation, then applies appropriate measures based on the assessed risk level.

Evolution Beyond Bot Blocking

Unlike traditional bot mitigation tools that often employ blanket challenges or blocking of all non-human traffic, Darwinium’s approach is more nuanced. It differentiates between legitimate AI agents and sophisticated threats like credential stuffing, content scraping, and inventory hoarding.

The platform offers a range of adaptive response actions. These include allowing low-risk agent interactions such as browsing and product discovery, conducting verification when behavioral signals suggest a shift, implementing step-up authentication to secure financial transactions or high-account risk scenarios, and blocking malicious automation attempting high-risk activities.

By applying controls at the edge, Darwinium ensures that enforcement happens before fraud can spread downstream, maintaining minimal latency. It also leverages emerging web standards, enabling AI agents to prove their identity through cryptographic signatures validated at the edge. This combination of techniques covers cases where automation does not self-identify as agentic.

Darwinium’s solution is designed to integrate with delegated transaction protocols like Visa Trusted Agent Protocol and Mastercard Agent Pay, along with emerging Know Your Agent models.

According to Alisdair Faulkner, the CEO of Darwinium, the key challenge lies in distinguishing between trusted automation and abuse while applying the right level of trust at the appropriate moment.