Forter Adds Fraud Controls for AI Shopping Agents

Forter is giving online merchants new controls for customers who send AI assistants to buy on their behalf. For ecommerce marketers, the immediate question is whether a legitimate customer’s order survives the same defenses that stop malicious automated traffic.

The company announced its Agentic Commerce Engine on October 13. The package brings fraud decisions together with checkout infrastructure and visibility into agent transactions. Fraud Management for Agentic is live for existing customers; Agentic Orchestration is available to interested customers.

Eran Vanounou, Forter’s chief technology and product officer, described merchants as:

“effectively opted in to agentic commerce”

A browser-capable assistant can attempt to use an existing storefront. A merchant still decides what traffic to admit and what evidence a purchase requires. Vanounou’s description captures the pressure to make those decisions before an unfamiliar automated order arrives.

A customer can arrive through software

An AI shopping agent is software carrying out tasks for a person, including finding a product and attempting checkout. Its speed and browsing behavior can make the visit look different from a human’s. Treating every automated session as abuse risks rejecting an authorized customer. Trusting every apparent assistant creates the opposite exposure.

Forter’s current product page describes an integration that connects merchants to several agent commerce protocols. Those are defined ways for software to exchange shopping and payment information. The vendor also offers account-based checkout intended to keep loyalty and customer recognition in the purchase flow.

That matters to a retail brand evaluating this channel. A purchase that reaches the till but loses the customer’s account relationship can be difficult to connect to a repeat purchase or a service request. Ask the provider to show which account and approval records remain available after the assistant finishes.

For a small B2B marketing team, the direct fit depends on whether customers buy through an online checkout. Teams selling through a sales conversation can use the same procurement question when an assistant acts on a customer’s behalf: what proof connects the requested action to that person’s authority?

Our earlier coverage of Shopify’s planned Muse checkout examined access to a purchasing route. Forter’s announcement concerns what a merchant does with the traffic once it reaches that route.

Identity needs to travel with the purchase

Checkout orchestration itself predates this week’s release. In a February 26 partnership announcement with Very Good Security, Forter already described its orchestration offering. October’s release brings that infrastructure together with the current fraud extension and activity view. Buyers should ask which capabilities are newly enabled in their own account.

The earlier partnership addressed the information lost as a payment travels through an assistant. A payment token substitutes a credential reference for sensitive card details. The companies proposed carrying Forter’s identity assessment alongside that token so the merchant receives more context.

Adam Davies, Forter’s Director of AI, said in that February announcement:

“Together, we’re ensuring that the risk signals captured during the Customer-to-Agent interaction persist downstream to the merchant”

The quote describes the partnership’s objective at the time. It doesn’t prove that every current agent transaction carries complete customer authorization. Davies’s public role and statement provide a useful question to bring to a demonstration: which signals reach the merchant today, and which remain with the assistant?

Approving the bot leaves another decision

Opus Research’s October 7 analysis separates agent identity from the scope of a person’s instruction. Knowing the assistant’s provider doesn’t establish permission to buy a particular item at a particular price. The analysis recommends retaining available approval evidence with the resulting transaction.

A merchant can test that distinction without promising a new revenue total. During a pilot, compare the final cart with the customer’s recorded approval. Check whether the authorization covers a substitute product or only the original item. Those are proposed acceptance tests, not incidents reported by Forter.

The company’s release says agent orders on its network grew more than fourfold since August. It doesn’t disclose absolute order counts, merchant mix, or detection definitions. That leaves the share of total ecommerce, and the amount of additional demand, unresolved. Forter has a commercial interest in encouraging merchants to buy its controls.

Marketing and fraud teams therefore need a shared pilot report. Count completed, approved orders and inspect the rejected ones. Keep the traffic classification separate from the evidence that the customer approved the purchase. The permission questions raised by Anthropic’s agent actions apply to customer-side automation too.

Before assigning an agent channel a revenue target, ask for one transaction record that shows the assistant’s identity, the customer’s purchase approval, and the completed order. Use the gaps in that record to define the next checkout test.