Your AI Can Find the Deal. The Store Still Controls Checkout.
![]() |
| AI-generated editorial illustration of restricted checkout access. This is not an actual retailer or AI-service interface. |
Asking an AI assistant to find the right charger is one thing. Asking it to buy the charger introduces several more participants: the store, the payment provider and whoever has to sort things out when the wrong item arrives.
That is where the promise of shopping agents is meeting resistance. An assistant can understand a request perfectly and still reach a checkout it is not allowed to use. Better reasoning alone cannot settle a disagreement over access to somebody else's shop.
Amazon's dispute with Muse makes the problem visible
Amazon confirmed to GeekWire in September that it had blocked Meta's Muse agent from shopping on its site. The company said the agent did not identify itself and raised concerns about how customer credentials were handled. These are Amazon's stated objections, rather than independently established findings about a security breach.
Meta's Muse launch explanation describes a different view of those protections: credentials are held in secure storage outside Muse's visibility, and the agent asks for approval before purchases. It also describes a separate permission system supervising the agent's actions.
Both positions matter. A safeguard within the assistant does not automatically mean a retailer has agreed to that assistant accessing its service. Equally, a retailer's objection does not prove that the assistant has exposed a customer's password.
Permission from the shopper is not the only permission involved
The current eBay User Agreement explicitly includes buy-for-me agents and other automated tools in its restrictions on accessing the service without eBay's prior express permission.
That wording is easy to misread. It refers to permission from eBay, not simply the shopper clicking an approval button in an AI app.
For someone trying to buy a product, the distinction can feel frustrating. You have chosen an assistant and told it what to do. From the store's perspective, however, automated access also raises questions about identification, traffic, fraud and responsibility for the transaction.
I see this as a coordination problem as much as an AI problem. The assistant, the buyer and the store need a shared understanding of what is authorised. A confident message saying the agent can handle everything is no substitute for that agreement.
The blocking figures need their dates attached
An audit published by Billions Network examined the public access rules of 51 major US retailers. It classified 37 as incidentally restricting shopping agents through existing automation rules, and found no dedicated purchase route meeting its definition of a verified, authorised agent checkout.
There are two essential limits to that finding. The audit examined published policies and robots.txt instructions, not proof that every attempted purchase was technically blocked. Its research was conducted in August, with sources checked through September 2, before Muse's launch and later partnership announcements.
Billions also sells identity infrastructure for agents, so this is research from a company with a commercial interest in the proposed solution. The dataset is useful evidence of fragmented rules; it is not a live census of every retailer or proof that AI shopping never works.
Some retailers are choosing partnerships
Meta's Connect 2026 announcement described additional shopping connections involving Walmart, Best Buy, Gap, Sephora and other retailers, alongside payment integrations. That points towards negotiated connections as one route forward, rather than unrestricted browsing being the only way an agent can shop.
The commercial stakes are easy to understand. A store that loses the browsing session may also lose opportunities to recommend related products and influence the purchase. An assistant that controls the shortlist gains a valuable position between the customer and the seller.
Those incentives help explain why access is strategically important. They do not establish that every security objection is a disguised attempt to protect advertising revenue.
A useful shopping assistant should make the limits visible
For me, the strongest version of this service starts with transparency: which shops were searched, which could not be accessed, what the final price includes and whether the product actually meets the requested specifications.
The purchase itself needs a clear approval point, a spending limit and an understandable route for cancellations or returns. These are the practical consequences of the permission questions around personal agents I covered earlier.
An assistant that compares options and hands over a reliable checkout can already be useful. The next advance will be getting the buyer, agent and retailer to agree on how the final step works.
.png)
Comments
Post a Comment