Blocking bots is no longer enough: AI agents now look exactly like your customers

The old plan: spot the bot, block the bot

For years, protecting a store from automation meant one thing: tell bots from people, and keep the bots out. Challenge pages, CAPTCHAs, browser fingerprints, "are you human?" checks. It worked well enough, because bots looked like bots.

What changed in 2026

AI agents don't need to look like bots anymore. Open-source tools, some with thousands of stars on GitHub, let an AI agent drive a real browser, the same Chrome your customers use. To your bot protection, it looks like a person: real browser, real clicks, real pages. Their authors report passing well-known challenge pages in testing.

Most people using them aren't attackers. They're customers who asked an assistant to "find my order and return it", and developers building those assistants. But the result for your store is the same:

• You can't tell an agent from a person anymore.
• The agent signs in with the customer's password, and does anything the customer could: buy, cancel, return, book.
• No rules apply to it beyond what a human gets. No limits made for automation, no approval of exact terms.
• Nothing records that an agent did it. When the customer says "I didn't ask for that", there's nothing to show.

And the more you tighten bot checks, the more you block the honest assistants too, while the determined ones keep getting through.

The question has changed

It's no longer "is this a bot or a person?". You'll lose that race. The useful question is:

> What is this agent allowed to do, for whom, and can I prove it afterwards?

That question has an answer, and it doesn't depend on catching anyone.

The answer: an official door for agents

Agents imitate people because often there's no other way in. Give them one, and the honest ones, which is most of them, will use it, because it's faster and more reliable than clicking through pages.

That's what Allowfy is: the official door for AI agents to your store or booking system.

1. Every agent is known. It proves who it is with its own key, or with a signature the way Visa, Mastercard and Cloudflare are standardizing. A key can't be faked by looking human.
2. The customer signs in with you, with a one-time code to their own email. The agent never gets a password.
3. Your rules decide every action: allowed, needs the customer's approval, needs your team, or refused, with limits per customer and per agent over time.
4. The customer approves the exact terms: "return order 1042 for 49", not "a return".
5. Every step is recorded on a signed ledger, with an evidence file per operation for any dispute.
6. Clear answers: refused agents learn why, so honest ones don't give up and send your customer elsewhere.

Be clear about what a door does and doesn't do

An official door doesn't stop someone from imitating a person on your normal website. Nothing fully does. What it changes is the incentive and the cost:

• Honest agents, most of the traffic, move to the door, where they're known and governed.
• What still comes through your pages is smaller, and easier to treat with suspicion.
• On your own site, keep the limits that matter whatever the visitor is: returns per customer, quantities per order, cash-on-delivery checks. Allowfy's per-customer limits work the same way on everything that comes through it.

The bottom line

Bot-blocking was built for a world where automation looked like automation. That world ended. You can't keep AI agents out, and your customers don't want you to. What you can do is let them in through a door you control: known agents, your rules, the customer's approval, and proof of everything.

Start free in test mode, and check whether your site is ready for agents with the free Allowfy Check.