Riskified Launches Tool to Detect AI Agents at Checkout

Riskified has launched Agent Identity Risk Intelligence, a new capability designed to detect when an AI-driven personal assistant — rather than a human shopper — is behind a transaction or customer service interaction. The product is built to flag autonomous AI agents, such as those branded "Muse," as they initiate checkouts or engage with support channels on a merchant's behalf.

The launch responds to the rise of agentic AI tools that can now browse, select products and complete purchases without direct human input at the point of action. Riskified, which has built its business on e-commerce fraud and risk decisioning, is positioning the new offering as a way for merchants to identify the source of a transaction — human or machine — before deciding how to treat it.

Agentic AI is moving from back-office automation into the final, trust-sensitive moments of commerce: checkout and customer service. Until now, most fraud and risk systems were built on the assumption that a person was behind every click, every payment detail entered, every support request. Agent Identity Risk Intelligence acknowledges that this assumption no longer holds, and that merchants need a new layer of visibility into whether a purchase or a service conversation is being driven by a human, a bot, or an AI assistant acting with delegated authority.

For leaders in digital transformation and AI adoption, this is an early signal of a broader shift: identity and risk infrastructure built for human behavior will need to be re-architected for a world where AI agents transact on people's behalf. It also raises operational questions — about liability, authentication and service design — that merchants, payment providers and platforms will need to work through together.

This launch is less about fraud prevention and more about a quiet redefinition of "the customer." When an AI agent can complete a purchase or raise a service ticket, the old binary of "verified human vs. suspicious bot" collapses, and businesses need a third category: authorized non-human actors operating with real purchasing intent.

Most operators will read this as a fraud-tooling update; the more important story is behavioral. Trust signals built for decades around human hesitation, friction and intent — mouse movement, typing cadence, time-on-page — mean little when an agent executes a transaction in milliseconds on a person's behalf. The real service-design challenge isn't catching bad bots, it's designing consent and accountability flows for good ones: making sure the human who delegated the purchase remains visible, reachable and responsible when something goes wrong. Operators who treat this as purely a security patch will miss the chance to redesign checkout and support for an era where the customer and the agent acting for them are no longer the same actor.