Elon Musk: Amazon Struggles to Distinguish Between Human and AI Buyers
Elon Musk responded to Amazon's ban on Meta's shopping agent Muse, stating that if access is through the user's own IP and cookies, Amazon will be unable to distinguish whether the order is placed by a human or an AI acting on their behalf.
The immediate trigger was Amazon cutting off Muse's ability to place orders on behalf of customers. A pop-up stated "Unauthorized AI agents continuing to access violate terms of use." Amazon accused Meta of not informing them in advance, not disclosing identity while browsing, and allegedly scraping stored user credentials; Meta claimed that the model does not see passwords and payment methods, and that credentials are securely stored for authentication. At the same time, Amazon also filed a complaint with the Ninth Circuit Court against Perplexity, alleging that its Comet copied Amazon session cookies to a cloud-based virtual browser for access.
This conflict arises at a time when agent shopping is just beginning to gain traction. Amazon itself has launched Rufus and Alexa for Shopping, with approximately 300 million users expected to have used Rufus by 2025, achieving a purchase probability of about 60%; CEO Andy Jassy publicly bets that customers will ultimately choose the retailer's own shopping agents over horizontal agents like ChatGPT. Amazon aims to protect recommendations, price comparisons, checkout, and advertising space, preventing third-party agents from turning the shelves into callable APIs.
On a technical level, Musk pointed out a vulnerability in session identity: if an agent resides in the user's browser, IP, and cookies, anti-scraping measures will still see "this account itself." Amazon is currently blocking identifiable automated browsers and undisclosed agents; once an agent fully simulates the local session, the cost of banning will turn into collateral damage against real users.
The buying side consists of companies like Meta, Perplexity, and those creating browser agents, betting that the agent layer will rewrite the e-commerce entry point; the selling side includes Amazon, which wants to maintain traffic and advertising revenue, and compliant funds that fear credential leakage. The event is strongly driven: on the day of the ban, the question of "who owns the shopping session" became a public confrontation. Beneficiaries are agents that can hide orders within user sessions; those under pressure are platforms that rely on identifying machine traffic to protect their shelves.
The market is trading the ownership of checkout rights: the shelves still belong to Amazon, but if decision-making power is seized by agents, advertising bidding and proprietary recommendations will depreciate.
Source: Public Information
ABAB AI Insight
Musk interprets the platform's ban as a failure of identity recognition rather than a victory of terms. This aligns with his consistent approach at X, xAI, and Tesla: bypassing gatekeepers through the distribution layer. During the PayPal era, the focus was on payment entry; the Twitter acquisition targeted information entry; this commentary addresses the e-commerce session entry. The historical controversy is that he has repeatedly claimed that "agreements cannot stop technology" as a given, with timelines often extended by platform counterattacks, but the direction is to let agents grow on the user side rather than on the platform's approved list.
The capital path involves agent companies using browsers, cookies, and local sessions to access retail inventory, transforming Amazon into a seamless fulfillment warehouse. Meta's push for Muse aims to connect advertising and social graphs to transactions; Perplexity uses Comet to capture the closed loop from search to order; Amazon, on the other hand, uses terms, pop-ups, and litigation to drive third-party agents out of the checkout funnel, leaving the incremental business to Rufus. The motivation is not the security rhetoric itself but to prevent shelves from being "freely accessed."
Comparative examples include Google blocking scrapers, the hotel industry blocking OTAs, and app stores blocking sideloading. OpenTable and Expedia once turned suppliers into interchangeable SKUs; now it's the turn of e-commerce shelves. The industry is in a phase of entry competition: retailers want their own agents, large model companies want general agents, and no party has yet achieved control.
The structural judgment pertains to the transfer of pricing power. The pricing power of advertising and recommendations is attached to "who is browsing." The mechanism is that when the buyer's identity shifts from an account to a session that can be reused by agents, the platform loses exclusive measurement of demand; banning can block identified robots but cannot stop agents that reside in cookies. The next round of charging rights will shift from product placement to the agent layer and wallet layer.
ABAB News · Cognitive Laws
Can block robots, but cannot block users' own sessions
Shelves remain, but if order rights are taken by agents, only fulfillment is left
Once identity can be reused by software, gatekeeping terms will depreciate