Amazon's AI commerce control: blocking agents, suing competitors, selling ads

In the week of September 21, Amazon fought AI commerce on three fronts. It blocked Meta's Muse shopping agent from Amazon.com. It filed a 41-page amended complaint against Perplexity, alleging its Comet for iOS browser copies user session cookies to Perplexity's own cloud servers and requests Amazon pages directly. And it confirmed a partnership with OpenAI to sell Amazon Ads inventory inside ChatGPT.

Three moves. One logic. Amazon is not reacting to AI commerce. It is defining the terms of access to its commerce data, its shopper relationships, and its advertising inventory. The platform that built the largest retail media network in the world is now deciding which AI agents may operate on its surface, which must negotiate, and which may buy their way in.

The Muse blockade

Meta launched Muse on September 8. The personal AI agent can browse, compare, and purchase across websites and marketplaces, connecting to Shopify or Stripe for payment. Twelve days later, Amazon cut it off.

GeekWire first reported the block on Sunday evening, September 21. Users who sent Muse to shop on Amazon began seeing a popup: "Continued access by an unauthorized AI agent violates Amazon's Conditions of Use, to which our customers have agreed." Amazon later confirmed to Bloomberg that blocking began Sunday evening.

Amazon's stated rationale: Meta didn't tell Amazon that Muse would access its store, the agent doesn't identify itself when it browses, and it appears to capture and store customer credentials. Meta disputes the credential-capture claim.

But the timing is revealing. Muse joins shopping agents from Google and OpenAI, both of which Amazon blocked over the past year. The pattern traces to a lawsuit filed against Perplexity in November 2025. Amazon is not blocking agents at random. It is enforcing a gate.

The gate is contractual. Amazon's Conditions of Use, updated with dedicated "Agent Terms" in May 2025, require AI agents to identify themselves, allow Amazon to restrict their access, and prohibit them from accessing Amazon after being told to stay out. The Muse popup cites the Conditions of Use, not the Computer Fraud and Abuse Act. That matters.

The Perplexity amended complaint

On the same day it blocked Muse, September 21, Amazon filed a 41-page amended complaint against Perplexity in the Northern District of California. The filing responds directly to the Ninth Circuit's August 4 decision vacating Amazon's preliminary injunction against Perplexity's Comet browser.

The Ninth Circuit held that the user, not Perplexity, was the one "accessing" Amazon's computers under federal anti-hacking law. The court denied rehearing on September 10. That ruling pushed Amazon to refile around contract and computer-fraud theories instead.

The amended complaint targets Comet for iOS, launched March 18, 2026, nine days after the district court injunction. Amazon alleges a specific architecture: Comet for iOS copies the user's Amazon session cookie to Perplexity's own cloud servers. Those servers run a virtual browser and request pages from Amazon directly. "No user device touches Amazon's servers," the complaint states.

That architecture contradicts what Perplexity told the Ninth Circuit. In briefs filed April 1 and May 6, 2026, Perplexity represented that "no Perplexity computer ever has direct access to an Amazon computer." Amazon says its engineers logged that traffic daily from March 18 through at least May 11. Amazon's counsel wrote to Perplexity's counsel on May 6, identifying each statement by page and asking for a correction. Perplexity repeated the statements in a reply brief filed later that day. Five days later, Perplexity's counsel wrote back: "As a measure of good-faith, Perplexity has temporarily disabled the assistant feature on Amazon.com in Comet for iOS."

The numbers Amazon is putting on it: damages well beyond $260,000. Eight members of its traffic engineering team have spent at least 1,280 hours on the agent since July 2025. As of June 15, 2026, Comet had run at least 185,712 sessions on Amazon.com. The complaint also returns to the user-agent string: Comet sends Google Chrome's. "Perplexity chooses that value, Perplexity codes it into the software Perplexity distributes," the filing reads. Perplexity has changed it each time Amazon learned to detect the agent.

This is not a copyright dispute. It is a fight over who controls the commerce session, who sees the shopper's intent data, and who gets to intermediate the purchase.

The ChatGPT ad partnership

While blocking third-party agents from its own surface, Amazon opened a different door. On September 10, Amazon Ads announced a partnership with OpenAI to support advertising in ChatGPT.

The integration allows participating advertisers to extend their Amazon Ads campaigns into a conversational advertising experience on ChatGPT. Select U.S. advertisers are currently testing the pilot. Delta Vacations is the first named brand, using the pilot to explore how consumer insights can inform when and how its offers appear in ChatGPT advertising experiences.

Amazon Ads says its advertising technology uses trillions of proprietary browsing, shopping, and streaming signals alongside supply relationships across Amazon properties and the open internet. OpenAI's ChatGPT reached 900 million weekly active users. Sensor Tower data shows positive performance for ChatGPT's ad network, with ads per user per hour of time spent increasing.

The partnership is asymmetric. Amazon brings commerce intent data and retail media infrastructure. OpenAI brings the conversational surface. Amazon sells the ads. The brands buying them are Amazon advertisers first.

This is not Amazon surrendering its surface to a third-party agent. It is Amazon extending its ad network into a surface it does not own, using data it does control.

Alexa for Shopping: the house agent

The third piece of the puzzle is the agent Amazon does allow: its own.

On May 13, 2026, Amazon retired the Rufus brand and merged it with Alexa+ into a single assistant called Alexa for Shopping, available on the Shopping app, the website, and Echo Show devices. By February 2026, Amazon disclosed that Rufus generated close to $12 billion in incremental annualised sales during 2025, with more than 300 million customers using the assistant across the year, monthly active users up 149 percent and interactions up 210 percent.

Research covered in April 2026 found the assistant compresses a results page of roughly 50 products into approximately five named recommendations per response. That is the new shelf. Five slots. No bidding.

At IAB Australia's Commerce & Retail Media Summit on July 7, 2026, Amazon's Jamie Jamison stated that consumers in the United States are now using the conversational chat agent more than they use the search bar. She described the interaction pattern: a shopper asks for a birthday present for a nine-year-old with a stated budget, or for the ingredients for a specific dinner, and the assistant performs the browsing, discovery, and review reading on the shopper's behalf.

Amazon's own consumer research, cited without published methodology or sample size, claims 75 percent of surveyed customers think about shopping multiple times a week, arguing that a discrete shopping occasion no longer exists as a planning construct.

Prime Big Deal Days, October 6-7, 2026, will be the first major event where Alexa for Shopping is the primary discovery surface. Acadia's Prime Day 2026 post-mortem found revenue rose 23 percent year over year across its clients, more than double the roughly 11 percent growth Adobe projected. Acadia also found that spend flowing to Amazon's sponsored prompts jumped 600 percent during Prime Day versus the immediately preceding period. Conversion showed which kind of AI interaction actually drove a purchase.

What this means for brands

The strategic picture is coherent. Amazon is building a controlled AI commerce ecosystem:

• Block third-party agents that access shopper data without permission, don't identify themselves, or operate through undisclosed architectures (Muse, Perplexity Comet, Google, OpenAI agents).
• Litigate on contract terms when anti-hacking law fails. The Ninth Circuit closed the CFAA door; Amazon opened the Conditions of Use door.
• Monetize external AI surfaces by extending Amazon Ads into ChatGPT, Perplexity, and wherever shopper intent concentrates, using Amazon's proprietary signals as the targeting layer.
• Own the house agent (Alexa for Shopping) that compresses the shelf to five recommendations, draws on product content, reviews, Q&A, and price history, and does not accept bids for placement.

The commercial implications are immediate:

Product content now feeds the agent. Alexa for Shopping and similar AI shopping tools draw on product content, reviews, customer Q&A, and other information when helping shoppers discover and evaluate products. PDP content, reviews, and Q&A now influence how products are surfaced and summarized by AI shopping tools. A brand that cleaned up its product attributes ahead of Prime Day gave the assistant better information to work with. Discount depth alone no longer wins the event.

Sponsored prompts are the new bid surface. Acadia's data shows spend on sponsored prompts jumped 600 percent during Prime Day. But the prompt types differ: product-level prompts trigger on specific ASIN queries, brand-level prompts fire on category or need-state searches. Both draw on product specifications, technical attributes, and customer review sentiment. The bid does not control the prompt. The evidence controls the prompt.

Measurement must travel with the data. ROAS from a retailer's onsite search report cannot carry the whole account once the campaign runs across AI agents, sponsored prompts, and external surfaces like ChatGPT. A useful scorecard needs to hold delivery quality, audience quality, commerce effect, and supply quality together.

Legal terms are becoming technical architecture. Amazon's Agent Terms in its Conditions of Use are now an enforcement layer. Brands and agents that ignore them face technical blocks. The Muse popup cites the Conditions of Use. The Perplexity amended complaint leans on tortious interference with contract. The ground has shifted to contracts.

The brands that adapt will treat Amazon as a platform with three distinct AI-mediated surfaces: the house agent (Alexa for Shopping), the paid agent surface (sponsored prompts), and the extended ad network (ChatGPT, Perplexity, and whatever comes next). Each has different economics, different evidence requirements, and different measurement.

The brands that do not adapt will keep optimising for a search bar that Amazon's own data says shoppers are using less.