What identity standards do shopping agents actually need?

Allglenn argues that Amazon blocking Meta's Muse from shopping on Amazon.com reveals a missing layer of commerce infrastructure: a standard way for a retailer to confirm an agent's identity, verify it is acting with the account holder's authorization, and enforce a refusal. Users who sent Muse to shop reportedly saw a popup stating that "continued access by an unauthorized AI agent" violates Amazon's Conditions of Use. The author frames Amazon's objection as having "three parts": Meta did not notify or obtain authorization before Muse accessed Amazon, Muse did not identify itself while browsing, and Muse "appeared to capture and store customer credentials and scrape account data."

The piece's reasoning is that Meta's countermeasures answer a different problem than the one Amazon raised. Meta describes a "Muse Secure VM" that vaults credentials so the agent model never sees a password, and a separate "Sentinel" agent that must approve actions such as sending an email or making a purchase. Allglenn credits this as a real consumer-security control but separates it from merchant-facing identification: "Secure storage is a consumer-security control. Self-identification is a merchant-facing protocol requirement. A system can satisfy the first and still fail the second." From this he derives five requirements a "durable agent-shopping interface" would need: agent identity attached to each request, verifiable user authorization, a merchant refusal the agent honors "rather than changing tactics," scoped credential boundaries, and defined commercial terms where an agent controls discovery or checkout. He places the dispute in a recurring pattern that also includes Amazon's suit against Perplexity's Comet browser and restrictions on agents from Google and OpenAI: "an agent uses a customer's credentials, the retailer says it did not authorize the agent, the agent does not identify itself, and the retailer blocks access."

This sits alongside Can a protocol split agent access from agent incentives?, which proposes an industry-built OAuth-based answer to nearly the same gap — consumer grants plus company-set limits. Allglenn's account shows what happens in the protocol's absence: lacking that machine-readable exchange, Amazon enforces identity and authorization through terms-of-use language and bot defenses instead, and a major platform relationship breaks down rather than negotiates. The dispute also complicates Can an AI agent serve both merchant and user interests fairly?: Azhar's loyalty conflict assumes Muse has already won shelf space on a retailer's site, but Amazon's block shows a retailer can refuse the agent outright before any referral-fee conflict comes into play — identity and authorization are a precondition a loyalty audit never reaches.

The account is built entirely from secondhand reporting (Business Insider, GeekWire, ADWEEK) on both companies' public claims, not from an independent technical audit of Muse's credential handling or of Amazon's access logs, so Amazon's allegations of credential capture and Meta's claims about Sentinel approval are each one side's characterization, not verified findings. The piece also does not establish whether Amazon's own "Buy for Me" tool — which Amazon says identifies itself and lets brands opt out — actually satisfies the five requirements the author lists, only that Amazon claims those two properties for it. If the pattern Allglenn describes generalizes beyond Amazon, agent operators that rely on an unidentified browser session, as Muse reportedly does, should expect further retailer blocks until an identity-and-authorization exchange, proprietary or standardized, is in place.