企业如何处理进入网站的 AI 智能体：拦截、转嫁风险或设置独立账户

An AI agent does things its customer would not, on that customer’s behalf. Three businesses answered that problem three ways in October.

People are starting to send AI agents to do things for them online: Buy the groceries, book the flight, check the bank balance. The agent goes to the business’s website and acts in the customer’s place. Every business now has to decide what to do when one shows up, and in October three big ones answered three different ways.

I think Walmart got it right: It is about to give each agent a separate profile inside the customer’s account. An AI agent will do things the person maybe wouldn’t have, so the two are not the same. But it does those things on the person’s behalf, so they need to be linked. A separate profile lets the business see both: the agent and the customer who sent it.

Chase Blocks AI Agents, Bank Of America Puts The Risk On You, And Walmart Is About To Give Them A Separate Profile

• Blocked. JPMorgan Chase CEO Jamie Dimon, on the Q3 2026 earnings call: “We are not gonna allow just any agent to come, outside agent to come inside the company.” The Verge reported that Charles Schwab blocks agents from accounts too, and that Delta is preventing them from booking tickets.
• The customer’s login, at the customer’s risk. A business that does nothing gets this: The agent signs in as the customer. Bank of America told customers never to share their login with an AI agent that offers to act for them, and that if they do, “you’re responsible for activity resulting from that access,” as TechCrunch reported. 1Password for Agents is built for this: an agent asks for a login, the person approves it, and “1Password then fills it directly into the sign-in form instead of sending it to the agent.” The release doesn’t say whether the website can tell an agent signed in.
• A separate profile. Walmart’s Agent Profiles, launching within a month of 21 October, give each agent “a dedicated profile within the customer’s account,” with approval for every purchase or “a spending limit they set – for example, $50, $100 or $200.” Walmart says agents see the same prices a person sees, and that agents that break its rules get blocked.

Walmart says why it built Agent Profiles: “Today, when a customer uses an AI agent, it signs in using a customer’s account. This means that all purchases – whether made by the customer or the agent – appear together, without any differentiation.”

An AI Agent Does Things The Customer Would Not, On The Customer’s Behalf

An agent might browse more pages than the customer would, maybe all of them. That’s the most basic difference. Beyond that, it could be literally anything.

So it’s not as simple as letting the agent sign in as the customer. Send a dog to shop for you at a local market, with a set amount of money. The dog only does it to bring the shopping back to you, and the whole thing only makes sense if the sellers know who sent the dog.

Walmart’s spending limit is the set amount of money, and the profile is the sellers knowing who sent it. A block stops the customer from using an agent at all. A shared login hides the agent behind the customer.

DoorDash Has Seen Several Hundred Agent Orders, So A Business Has Time To Decide

DoorDash says outside AI agents have placed “maybe several hundred orders in total”, in CEO Tony Xu’s words on the Q3 call, as TechCrunch reported. In the same quarter DoorDash handled 993 million orders, more than 10 million a day.

The rules for how an agent tells a business who it is are still a draft. The AI customer-service company Sierra and Meta are writing them as the Personal Agent Protocol: 40-plus companies are helping shape it, and its website named no business using it with customers when I checked on 1 November. On October 30, the draft added a recommendation that agents sign their requests with Web Bot Auth, a proposed way for an agent to prove which company runs it. According to the bot-protection company DataDome, as PPC Land reported, Meta’s Muse agent has been signing some of its requests that way since around 12 October, without an announcement.

Do three things before agents arrive in numbers:

1. Check what the software that keeps bots off your website does to an AI agent today. It might block it, let it through, or be unable to tell it from a person.
2. Decide which of the three you would choose, and write it down, before your platform or your security vendor decides for you.
3. Ask whoever runs your website to watch for agents that sign their requests. Those are agents telling you which company runs them.

If you can’t tell the agent from the customer, you can’t treat it differently.

More Resources:

• Google-Agent: The Web’s New Visitor Just Got An Identity
• Your Store Is ‘Agent-Ready’ By Default, We Tested 122 Shopify Stores To See What That Actually Means
• Shopify Now Lets Stores Pay The Company Behind Your AI Shopping Agent

This post was originally published on No Hacks.

Featured Image: Mr.paripat niyantang/Shutterstock

Slobodan Manic Founder of No Hacks and machine-first website optimisation consultant at No Hacks

Slobodan “Sani” Manić is a website optimisation consultant with over 15 years of experience helping businesses make their websites faster, ...