Recognizing Good Agents: Signals and Strategies Required to Reduce Customer Insult

The commerce space has evolved rapidly over the last year, where we see an increasing set of experiences wherein agents assist customers in discovery, cart building, and purchase of products across merchants online. Shoppers get more choice and a familiar flow, while merchants gain access to high-intent demand without forcing users to jump between sites. But this innovation also introduces a new attack surface. As agents gain autonomy and operate across merchants, compromised or malicious agents can scale fraud faster while obscuring intent and liability. At the same time, agentic commerce also challenges the foundations of fraud prevention. When agents run in data centers, traditional signals such as IP reputation, device fingerprinting, cookies, and session continuity go missing. Bot defenses mistake legitimate sessions for attacks, and fraud engines lose the ability to see and connect activity to the principal, the human whose intent and liability underpin every transaction. The result is more CAPTCHAs, more false positives, and increased operational costs, just as consumers are being introduced to agent-based shopping for the first time. This session unpacks the risks, shows where current models break, and highlights why adoption hinges on keeping merchant integration as close to zero-lift as possible. We will present strategies that allow agents to be recognized as trusted actors. Key Takeaways: The fraud challenge - Data center agents break traditional trust checks and weaken principal visibility, driving more false positives and friction. Why the principal matters - Risk needs a verifiable link to the human behind the agent, not just the agent itself. Adoption reality - Solutions that force merchant teams to rewrite code stall. We outline a low-lift direction that keeps merchant changes minimal while improving trust signals. The new attack area and velocity introduced by agents gaining more autonomy and reach.

Some content is hidden, to be able to see it login here Login