Merchants distinguish legitimate AI shopping agents from hostile bots through agent authentication: the agent presents a verifiable identity credential, typically a cryptographic signature on its requests that traces back to a known operator, and the merchant checks it before granting access. Behavioral bot detection alone cannot make this distinction, because a well-built buying agent and a scraper look nearly identical on the wire. The emerging framework for this is KYA, Know Your Agent, an identity-verification layer for software actors modeled loosely on KYC for customers.
The reason this problem is hard is that the old defenses were built on one assumption: automated traffic is bad traffic. CAPTCHAs, IP reputation, and browser fingerprinting all exist to keep non-humans out. Agentic commerce breaks that assumption. An AI agent arriving to buy something with a real customer's money is exactly the traffic a merchant wants, and it is automated by definition. A merchant that blocks all bots blocks its newest customers. A merchant that unblocks all bots invites scrapers, card testers, and inventory hoarders in through the same door.
Agent authentication resolves the tension by moving the question from "is this a human?" to "who operates this agent, and can they prove it?" The mechanics as of September 2026 generally involve signed HTTP requests: the agent's operator holds a cryptographic key, publishes the public half where merchants can find it, and signs each request so the merchant can verify origin. Search infrastructure has worked this way for years, which is how sites confirm a crawler claiming to be a search engine actually is one. The same pattern is now being extended to commerce agents, with proposals for standard request-signing schemes working through web standards bodies.
Identity is the first check, not the whole answer. A verified agent can still misbehave, so merchants layer on what the agent is allowed to do: rate limits, purchase caps, and access scoped to product data rather than the whole site. Some operators also verify the human behind the agent, so a purchase carries a chain from customer to operator to request.
The unresolved part is coverage. Signing schemes only help with agents whose operators participate, and the long tail of agents will not. That leaves merchants running two systems side by side: cryptographic verification for the known operators, and the old behavioral arms race for everyone else.
Go deeper: the merchant agent-authentication problem and how access control splits agents from crawlers.