Amazon has blocked access to Meta's personal AI agent, Muse, on its shopping platform. Following similar actions by Perplexity, OpenAI, and Google, this move solidifies the trend of data-rich platforms restricting free access to their 'data moats' for AI agents.
According to the IT industry on September 26, Amazon began blocking Muse's access to Amazon.com on the night of September 20. After Amazon requested Meta to remove Muse, which Meta declined, the block was implemented. Users attempting to shop on Amazon through Muse now receive pop-up messages indicating a violation of terms of service.
Muse, launched by Meta on September 8, is a consumer AI agent that handles tasks such as travel bookings, email and schedule management, and payments. It is available on apps, the web, and WhatsApp, with both free and paid subscription options priced at $20 and $100 per month. Despite quickly rising to the top of the app store, Muse found itself blocked from the largest online shopping site in the U.S. within two weeks.
Amazon cited security concerns as the reason for the block, claiming that Meta did not provide prior notice and that the agent did not disclose its bot status while accessing the site. Additionally, Amazon alleged that the agent appeared to collect and store customer login information. Meta has countered these claims, with Tarek Shehata, Vice President of Meta's Superintelligence Labs, explaining that login information goes directly into a secure storage area and is not visible to the main agent.
This block is not an isolated incident. Over the past year, Amazon has sued Perplexity over the Comet browser and has also blocked shopping agents from Google and OpenAI. In the case of Google, Amazon modified its robots.txt file to restrict crawler access, resulting in the removal of Amazon product pages from some AI search results. All blocked entities lack a strong or independent commerce base, effectively shutting down a structure that relied on borrowing data from other shopping sites.
The underlying motive appears to be advertising revenue. Amazon generated over $68 billion from advertising last year, a business model that relies on users browsing pages and viewing sponsored products. In the second quarter of this year, advertising revenue reached $19.81 billion, a 26% increase from the previous year. If agents select products and complete purchases, all points of interaction where Amazon earns revenue—such as search bars, comparison screens, and payment flows—would disappear.
Meta has previously employed similar reasoning. In October of last year, Meta changed its WhatsApp Business terms to block access for general AI chatbots like ChatGPT and Perplexity, with the new rules taking effect on January 15. Amazon is not free from double standards; its 'Buy for Me' feature accesses external shopping sites to input names, addresses, and payment information for items not available on Amazon. This structure allows its own agents to use others' data while blocking access for competitors, raising questions about data ownership rather than principles.
Legal avenues are also becoming less effective. In August, the U.S. Ninth Circuit Court of Appeals weakened Amazon's strategy under the Computer Fraud and Abuse Act (CFAA), leaving enforcement of terms of service as one of the few remaining options. As a result, technical blocks and negotiations are likely to become the primary battleground.
Industry experts predict that data access will increasingly become a paid service. This could involve big tech companies forming 'data gentleman's agreements' or agent developers paying fees to data holders. For instance, Google has established a $60 million annual data licensing agreement with Reddit, setting a precedent in the content space. The competition among agents is shifting from model performance to securing retail, booking, and content data.
However, not all companies with data will hold the upper hand. Negotiating power stems from the exchangeable nature of data, which only a few companies possess. Many firms either do not recognize the value of their data or have not refined it into 'AI-ready data' that can be immediately utilized. This highlights the need for companies to focus on data management rather than merely implementing models as they transition to AI.
South Korea is entering a similar landscape. Naver holds data assets in search, shopping, and blogs, while Coupang has data on products, prices, and delivery, and Kakao possesses content from its Daum portal. As global agents begin to enter the domestic market, the access policies of these platforms will set the baseline for negotiations.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.