On 8 September Meta launched Muse, a personal AI agent that can shop across the web on a user's behalf, checking out with Link by Stripe, which issues a single-use card for each approved purchase. It is rolling out in the US only. Within two weeks Amazon had asked Meta to remove its store from Muse and, when Meta declined, blocked the agent, Retail Gazette reports, telling users that access by an "unauthorised AI agent" breached its terms. An Amazon spokesperson told Retail Dive that agents buying on customers' behalf "should operate openly and respect service provider decisions about whether or not to participate". Amazon is separately suing Perplexity over its Comet agent.
Muse is not in Australia yet, but the question it forces is coming. Adyen's 2026 research found 64 per cent of Australians have already used AI assistants to browse, compare or discover products, even though 64 per cent are uncomfortable letting AI complete the purchase. Discovery through agents is already normal here. When checkout follows, every Australian retailer with a website will be making Amazon's decision, whether it means to or not, because doing nothing is also a choice.
Blocking agents is the wrong reflex. Amazon can afford it: it is the destination, and it has its own assistant to route customers to. Almost no Australian retailer is in that position. If an agent is asked to find a pair of running shoes under $200 and your site turns it away, it doesn't give up. It buys from a competitor that let it in.
But letting any agent in on any terms is also wrong, and this is where Amazon has a point worth taking. The complaint was not that Muse was an agent. It was that Muse arrived unannounced, didn't identify itself and handled customer logins in a way Amazon hadn't agreed to. That is the line I would draw: don't block agents, block anonymous ones. An agent that identifies itself, transacts through a checkout you have sanctioned and leaves you as merchant of record is a new channel. An agent that scrapes your pages and drives your checkout pretending to be a person is a risk to your fraud controls, your pricing and your customer relationship.
The practical difference is the protocols. The standards covered in our guide to agentic commerce exist precisely so an agent can declare itself and a merchant can decide what it may do. Retailers that expose an agent-ready checkout get to set the rules. Retailers that don't are left choosing between a blanket block and being scraped.
Write down your agent policy before you need it: which agents you will serve, what they may do (browse, add to cart, buy), whether you will accept agent logins at all, and how a disputed agent order is handled. Then ask your commerce platform and payment provider when they will support identified agent checkout in Australia. The retailers who answer those questions early will be the ones agents are allowed to recommend.
Get the latest thinking on AI-powered retail — from product personalisation to in-store innovation — delivered to your inbox once a month.