Why eBay Drew the Line at AI Checkout

eBay will begin blocking third-party AI “buy-for-me” agents from autonomously placing orders on its platform starting February 20. The change appears in an update to eBay’s User Agreement, which explicitly prohibits “buy-for-me agents, LLM-driven bots, or any end-to-end flow that attempts to place orders without human review” unless the tools are approved by eBay.

eBay is drawing a clear boundary around who is allowed to automate commerce on its platform, and under what conditions. Like most platform rules, enforcement is imperfect. The aim is not to eliminate automation entirely but to restrict it at scale through account controls, contractual leverage, and selective access, rather than allowing unrestricted autonomous buying to proliferate unchecked.

This is not a rejection of AI. eBay continues to deploy AI internally, including AI-assisted messaging tools for sellers and other automation features highlighted in recent earnings calls. What it is rejecting is uncontrolled, third-party automation at the point of transaction. That distinction (between AI that assists users and AI that executes purchases) is now shaping policy across the industry.

AI Can Browse. Checkout Is Off-Limits.

eBay’s decision mirrors moves already underway at other large commerce platforms, most notably Amazon. Amazon has taken steps to block external AI shopping agents from accessing its site, including technical restrictions and legal action against companies it claims are using automated agents without authorization, even as it develops its own agentic features such as Rufus and experimental “Buy for Me” tools.

The contradiction is only apparent. Amazon’s posture, like eBay’s, is not anti-agentic commerce. It is anti-external control. Amazon’s financial disclosures help explain why. In 2024, the company reported more than $156 billion in revenue from third-party seller services and over $56 billion from advertising, both of which depend on controlling the shopping funnel, pricing signals, and attribution. Autonomous agents that bypass Amazon’s checkout threaten that control.

eBay faces similar incentives at a different scale. The company reported roughly $75 billion in gross merchandise volume in 2024 against about $10.3 billion in net revenue, implying a take rate in the low-teens. Small shifts in pricing efficiency or buyer behavior can materially affect that model, particularly in auctions and competitive fixed-price listings.

Automation itself, however, is not new to marketplaces. Tools for proxy bidding, price tracking, and last-second sniping have existed on eBay and similar platforms for years. What has changed is scope. Newer agents don’t just optimize bids or timing. They select items, negotiate prices, and complete purchases end to end, often without the user reviewing listings, descriptions, or terms.

This is why marketplaces are drawing a sharp line between discovery and execution. AI-driven discovery is already widespread. Consumers routinely use chatbots and search tools to research products and compare prices before clicking through to a retailer. What platforms are resisting is AI that completes the transaction independently, outside their systems of measurement, safeguards, and monetization. Like most platform rules, enforcement is imperfect, but the intent is clear: restrict automation at scale through account controls, contractual leverage, and selective access rather than attempt to eliminate it entirely.

Agentic Commerce Is Becoming Gated

Early visions of agentic commerce imagined a single AI agent shopping seamlessly across the web on a user’s behalf. That vision is already colliding with platform reality. Instead of openness, the industry is moving toward permissioned access.

One response has been to build formal rails rather than allow ad-hoc automation. OpenAI launched Instant Checkout in late 2025, enabling purchases inside ChatGPT for partners such as Etsy and selected Shopify merchants, with merchants opting in and paying a fee on completed transactions.

Google and a group of retail partners, including Walmart, Target, Shopify, and Etsy, announced the Universal Commerce Protocol earlier this year with similar goals: allowing AI agents to transact through standardized, authenticated interfaces rather than uncontrolled browser automation. Agentic commerce is welcome, but only through defined infrastructure that allows platforms to manage access, risk, and economics.

Safety arguments continue to dominate public explanations, and they matter. Blocking abusive bots, fraud, and unfair competition is necessary for marketplace trust. It also creates legal distance between platforms and autonomous systems that may mis-execute purchases, trigger disputes, or act outside a user’s intent, an outcome that could otherwise increase chargebacks, arbitration, and seller-buyer conflicts.

The web remains open, but the rails that move money increasingly are not. Autonomous agents collapse browsing, comparison, and impulse into a single step, reducing time on platform, secondary purchases, and the behavioral data marketplaces rely on to price listings and sell ads.

The likely outcome is fragmentation. Instead of one universal shopping agent, consumers will encounter different agents tied to different ecosystems. AI builders will be forced to partner or integrate rather than scrape or simulate human behavior. Merchants will face a choice: opt into permissioned agent rails or risk being invisible as buying behavior shifts.

eBay’s move clarifies the terms under which agentic commerce will exist. Across the industry, platforms are signaling that automation at checkout is too strategically important to leave unmanaged.