Aidō Lighthouse
Get a demo
Resources / AI agent trust
Commentary · 14 Oct 2026 · Updated 14 Nov 2026 · 7 min read

Who does a business trust when an AI agent arrives?

A personal agent can search for a product, compare options and place an order. When it reaches a retailer, the retailer has to decide whether to let it proceed. What tells the business who is acting and what the customer approved?

Meta’s Muse has been shown using Poppy to work with a business agent. JPMorgan has warned about letting outside agents into customer accounts. American Express published a framework that records purchase intent before payment, while Forter announced tools to identify agent traffic and manage checkout. For merchants, trust decisions now reach from access and authorization to the purchase itself.

Poppy 1.0 gives agents and businesses a shared starting point

Personal agents often navigate sites through pages built for customers. The business may see automated requests without knowing which software is behind them, who the agent represents, or what the person approved.

Sierra announced Poppy 1.0 on 13 November 2026. A company can publish how a personal agent should use its website, APIs or company agent. Sierra also reports that Shopify now publishes Poppy discovery for every merchant and that Meta’s Muse checks for Poppy first. Those platform announcements do not establish that every route works on an individual storefront.

The 1.0 specification supports gradual adoption: a business can start with a website, an API or an agent. OAuth metadata connects its domain to the right issuer; scope descriptions explain the account access an agent can request. A published discovery file establishes declarations, not proof that a customer can complete a task.

What changes for merchants

The UCP extension connects Poppy identity and authorization to commerce capabilities, including catalogue search, carts, checkout and UCP Ask. Declaring it does not demonstrate successful search, consent enforcement or payment.

What Aidō reports

Aidō’s Poppy checks cover discovery structure, public OAuth metadata, declared API descriptions, browser controls and extension declarations. Reports distinguish functions listed in an OpenAPI description from operations actually executed. Lighthouse, GEO, vendor scans and ARIA’s Muse simulation share this discovery path.

These are read-only checks. They do not connect to Muse, sign in, create a Poppy session, fetch session-dependent skills, submit feedback, or perform checkout. Capability coverage records those limits so a protocol declaration is not presented as a completed transaction.

What the recent launches cover

The announcements address different parts of the transaction:

01

Know what is asking

JPMorgan CEO Jamie Dimon said the bank would not let just any outside agent into customer accounts. He warned that customers need to understand the risks of giving agents access to their data.

02

Carry the customer’s intent

American Express’s Intent Intelligence framework records details such as the merchant and maximum budget, then asks the customer to authenticate and approve them. A Proof of Intent Token can help scope the payment credential used by the agent.

03

Give merchants visibility and control

Forter announced tools for identifying agents, supporting headless checkout and showing merchants which agents transact. The company says agentic orders across its network grew more than fourfold since August. That figure is Forter’s own network data.

Poppy describes how an agent and a business can interact. Amex is working on how customer intent informs payment. Forter focuses on merchant trust decisions and checkout. Each addresses a different part of the transaction, and none makes a purchase safe on its own.

A trusted agent still needs a usable storefront

Even with an identifiable agent and an approved task, the merchant’s site has to support the journey. The agent needs to find the product and read its price, availability and terms. The site also needs a usable path through to the next step, without an access control blocking legitimate traffic.

Aidō Lighthouse looks at the merchant side through read-only scans of public storefronts. We assess Discoverability, Understandability and Transactability: whether agents can find products, interpret the information and encounter usable paths toward purchase. Scans also report access signals such as bot protection and challenge pages, which can show where automated traffic meets friction.

A scan cannot authenticate a shopper, validate an intent token or certify an agent as safe. It records evidence from the public storefront: what the scan could reach, what product information it could interpret and where the tested path stopped. Teams can use that evidence to decide what to improve and what access rules to apply as protocols develop.

“Customers decide what access to give their personal agents, and companies set parameters for what those agents can do.” That is the Poppy principle.

As protocols like Poppy develop, storefront scans give merchants concrete evidence to work from.

See what an AI agent can do on your store

We scan e-commerce sites to show what AI agents can and cannot do. Get in touch to discuss a scan.

Get in touch