Riskified has launched a service to help merchants assess the people and behaviour behind AI shopping agents, with a limited beta planned for December. The move brings fraud checks into the emerging world of assistants that can place orders and request refunds for customers.
Watch Desk analysis
What happened
Riskified says its Agent Identity Risk Intelligence can identify when an order or customer-service request comes from a personal AI assistant, connect it to the person behind the agent and return a risk signal. The company says the service can use verified agent credentials when available, alongside identity, network history and agent behaviour. Read Riskified’s announcement.
The service is due to enter a limited beta for enterprise merchants in December 2026, with general availability expected in 2027. Riskified says it will first be available to merchants on Shopify, where orders placed through Meta’s Muse already arrive tagged as agent-originated. The company also plans reporting by assistant, policy controls by agent type, and risk information in its Zendesk integration.
Our top picks
- Identity signals for agent-led orders
Riskified says it can connect an agent’s request to the shopper behind it, even when browser and device signals are missing. - Agent-specific merchant policies
Decision Studio will let merchants set different rules for different agent types and risk levels. - Reporting by assistant
Control Center will show agent-originated orders and claims, with breakdowns by assistant where identifiable. - Shopify is first in line
The initial rollout will cover merchants on Shopify, where Muse orders are already tagged as agent-originated. - Risk information in Zendesk
Risk indicators are intended to reach both human support staff and Zendesk’s specialised AI agents handling refund or return requests.
Why it matters
An authorised agent is not automatically a trustworthy shopper. Riskified argues that valid credentials alone cannot rule out a compromised account or an agent being used to automate abusive claims. Its proposed system is designed to help merchants distinguish between the agent, the person authorising it and the behaviour in a particular transaction.
That is a practical problem as assistants take on more shopping and customer-service tasks. Riskified’s announcement describes a commercial product, not evidence that its risk signals will identify fraud accurately or improve outcomes for merchants. The planned beta will be the first meaningful test.
Our read
This is a timely piece of infrastructure for agent-led commerce: if assistants can buy and seek refunds, merchants will want to know who they are acting for. The interesting question is whether agent-specific risk checks can protect merchants without turning legitimate shoppers’ automated errands into a fresh obstacle course. Riskified has put a product and a timetable on that question; the beta results matter more than the pitch.
What to watch
- Which enterprise merchants join the December beta and what results they share.
- How Riskified distinguishes legitimate agent behaviour from abusive activity.
- Whether the service expands beyond Shopify as agent-identification standards take hold.
- When general availability arrives, and which capabilities are included.
Discussion spark: Should merchants be able to apply different rules to AI shopping agents than to the people authorising them, or should an agent simply inherit its user’s trust?
Sources and evidence
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.