Riskified launches agent identity tool as AI shoppers blur fraud lines

Agentic checkout assistants are rewriting retail fraud detection, forcing merchants to ask not "is this a bot?" but "who authorised it?"

By Disrupts Editorial Team Fri, 16 October 2026 AI level 2 of 5

AI level 2 of 5: drafted by our AI editorial assistant from the company's announcement; fact-checked, edited and signed off by Mark Walker, Editorial Director. What the levels mean

Five identical smartphones on clear display stands are arranged on a white counter in a brightly lit tech showroom, with blurred shelves of products and plants in the background.

Riskified, the NYSE-listed fraud and risk intelligence platform, has announced Agent Identity Risk Intelligence, a product designed to identify the consumer AI assistants now placing orders, filing refund claims and initiating chargebacks on shoppers' behalf, and to assess the trustworthiness of the person who authorised them. The launch arrives as agentic commerce shifts from novelty to mainstream retail infrastructure, with Meta's Muse, Sierra's Instinct and OpenAI's ChatGPT Dots already capable of completing end-to-end purchases without a human touching a keyboard.

The timing is pointed. Riskified's own UK consumer research found that 37.9% of British shoppers would already delegate a refund or credit complaint to an AI agent, and 23.4% would use one to initiate a bank chargeback if a return proved difficult. Behaviours that were occasional for human shoppers, price-drop refund hunting, return-policy arbitrage, cart abandonment on a better offer, become persistent, always-on routines when an agent executes them at machine speed across every order.

The authorisation gap that new protocols cannot close

The retail industry is already moving to standardise agent authentication. The Personal Agent Protocol, introduced this month by Sierra and Meta alongside Shopify, Stripe and Walmart, gives merchants a cryptographically secure way to confirm that an AI assistant is genuinely authorised to act on a given account. Riskified says it supports the standard, and will accept verified agent credentials where they exist. But it argues the protocol solves only the first risk question.

An agent authorised by a compromised account is still an account takeover. An agent authorised by a synthetic identity is still a first-order fraud. And an agent authorised by a refund-as-a-service fraud ring, whose members hold real accounts and grant real permissions, is organised crime running at software speed through support channels built to serve legitimate customers. The authorisation layer confirms the agent is permitted. Riskified positions itself as the layer that determines whether the person who issued that permission should be trusted at all.

The threat is not hypothetical. In April 2026, eleven members of the Artemis Refund Group were sentenced in a US federal court after running refund fraud as a subscription service, placing thousands of fraudulent claims against Amazon, Walmart, Target and Wayfair. That operation was manual. Add an autonomous AI agent to a similar ring and the scalability of claim abuse changes fundamentally.

Cross-sector implications: from retail checkout to the fraud economy

"Agent protocols solve a real problem: merchants need a secure way to let a customer's assistant act for them. What they don't solve is whether that customer should be trusted," said Assaf Feldman, Chief Strategy Officer and Co-Founder of Riskified. "A fraud ring can authorise an agent as easily as a loyal shopper can."

The convergence at play here cuts across several sectors simultaneously. For retail and e-commerce, the immediate pressure is operational: fraud detection logic built on device fingerprints, browser telemetry and session history is structurally blind to agents that check out headlessly and abandon no cookies. Riskified's response is to lean on identity and network history drawn from billions of orders rather than device signals.

For the broader financial services and payments infrastructure, the emergence of agentic commerce raises harder questions about liability. When an AI agent initiates a chargeback on a customer's behalf, who bears responsibility if the claim is fraudulent? Card networks, acquiring banks and payment processors are watching this space closely, with no settled regulatory framework yet in place.

For investors tracking the fraud-prevention and identity verification market, the Riskified launch reflects a broader capital thesis: that the shift to agentic workflows will force every trust-and-safety stack to be rebuilt from the identity layer up. The company's AI Agent Intelligence suite, first introduced in August 2025, now extends to post-checkout service interactions via a Zendesk integration and is available through an MCP server on AWS Marketplace. A limited enterprise beta opens in December 2026, with general availability expected in 2027. The race to define the standard for agent identity risk is effectively now open.