Here is a statistic from 2026 that should change how you treat your logs: when researchers checked traffic that identified itself as a well-known AI agent, a majority of it came from addresses that did not belong to the operator it named. In other words, most "AI agent" traffic hitting sites is impersonation — ordinary bots wearing a trusted name because they have learned that a friendly user-agent string gets waved through. As AI agents become real customers, this impersonation becomes a real attack, and it is one you can shut down cleanly.
Why a user-agent string proves nothing
A user-agent is just text the client sends. Anyone can set it to anything. A scraper can call itself a mainstream browser, a search crawler, or the latest AI shopping agent, and nothing about the string itself is verifiable. Allow-listing by user-agent — "let the good bots through" — is therefore an open door: you are trusting a label the attacker writes. The real crawlers and agents you want to trust all have something the impersonators cannot fake: they operate from their own published networks.
Verify the claim against the network
Legitimate operators run their crawlers and agents from infrastructure they own, announced under their own Autonomous System Number (ASN) — the identifier for the network a given IP belongs to. The verification is straightforward:
- Take the IP the request actually came from.
- Look up its ASN — the network that truly owns it.
- Compare that to the operator the user-agent claims to be. A genuine request from a named crawler resolves to that company's network. An impostor resolves to a residential proxy, a random hosting provider, or an unrelated network entirely.
When the declared identity and the real network disagree, you are not looking at the bot it claims to be. That one check turns an unverifiable label into a verifiable fact.
Where this matters most
- Agentic checkout: deciding whether to trust an AI shopping agent with a purchase.
- Content and pricing protection: telling a legitimate search crawler from a scraper harvesting your catalogue behind a trusted name.
- Rate-limit and allow-list hygiene: making sure your "trusted bot" exceptions cannot be borrowed by anyone who copies the string.
Make the network the source of truth
Fraudex resolves any IP to its network and ASN, with a type and bot verdict, in a single call — so you can verify a declared agent or crawler against the network it actually came from instead of the name it typed. Look up a network directly with the ASN lookup, screen requests with the proxy detection API, or spot-check an address with the IP checker.