E-commerce Platforms Lock Out AI Shopping Agents Despite Industry Push

Retailers Draw the Line

eBay has implemented restrictions preventing AI agents from accessing its marketplace, marking another setback for technology companies promoting automated shopping as the next frontier of consumer behaviour. The platform now blocks tools including Meta's Muse agent from retrieving product listings, pricing information, and other marketplace data.

A user who had configured Muse to compile daily summaries of watch listings found the service unable to retrieve any eBay content in recent weeks. The agent reported being blocked from the site entirely. In some instances, Muse redirected users to external websites that republished eBay listings rather than linking to the platform directly.

The restrictions represent a practical obstacle for an industry narrative that has positioned AI agents as transformative tools for online shopping. While technology firms describe a future where automated systems book travel, compare prices, and complete purchases without human intervention, the platforms hosting those transactions are implementing barriers that call the viability of that vision into question.

eBay had previously modified its user agreement to prohibit "buy-for-me agents" and "LLM-driven bots" that execute transactions without human oversight. Until recently, however, agents retained the ability to browse listings and aggregate information for users, a capability that has now been curtailed.

The Data Access Question

Sybille St. Arromand, an eBay spokesperson, confirmed that the company takes a selective approach to third-party access. The platform has restricted agent activity in cases involving unauthorised data scraping and training, she said, whilst declining to specify whether new limitations had been imposed in recent weeks.

The company is evaluating how third-party agents might help new customers discover eBay's inventory, St. Arromand said, with a focus on preserving the trusted experience that buyers and sellers expect. The statement suggests eBay views agent access as a partnership question requiring formal agreements rather than an open capability.

That stance reflects a broader tension over who controls the commercial relationship with shoppers. E-commerce platforms have spent years refining recommendation algorithms, checkout flows, and personalisation features designed to maximise conversion rates and customer lifetime value. Allowing external agents to mediate that relationship introduces uncertainty about attribution, data ownership, and revenue sharing.

The issue extends beyond technical access. Retailers invest heavily in brand experience, from curated homepages to loyalty programmes that encourage repeat visits. An agent that strips away that context and presents product information as commoditised data threatens to erode the differentiation retailers have built. For platforms like eBay, which earn revenue from listing fees and transaction commissions, maintaining direct relationships with both buyers and sellers is essential to the business model.

A Fragmented Landscape

The approach to AI agents varies considerably across e-commerce. Amazon has blocked external agents entirely, preventing them from browsing or transacting on the platform. Target enforces restrictions that allow agents to succeed in some browsing scenarios whilst blocking them in others, creating an inconsistent experience. Shopify has taken the opposite approach, opening access to its catalogue of more than one billion listings so that shoppers can configure agents to complete purchases on their behalf.

This fragmentation creates a challenging environment for both developers building agent platforms and consumers attempting to use them. An agent that functions seamlessly on Shopify may be entirely blocked on Amazon, rendering it unreliable as a general-purpose shopping tool. The lack of interoperability standards means each retailer can set its own rules, and those rules can change without notice.

The inconsistency also complicates the value proposition for agent platforms. If a tool can only access a subset of retailers, it becomes less useful as a comparison-shopping mechanism. Users seeking the best price or selection may find that the agent cannot reach the platforms with the most competitive offers, forcing them back to manual browsing.

Security and Abuse Vectors

Retailers cite legitimate concerns about bot activity beyond simple data scraping. An incident in October illustrated how agents could be weaponised to disrupt marketplace operations. A user demonstrated that Muse could repeatedly add items to a shopping cart on The RealReal, a luxury consignment platform, effectively locking out other buyers. The RealReal subsequently updated its bot detection system.

That scenario highlights a fundamental mismatch between retail infrastructure designed for human shoppers and the capabilities of automated systems. E-commerce platforms assume constraints on user behaviour: a person can only browse so many pages per minute, add so many items to a cart, or monitor so many listings simultaneously. Agents operate without those limitations, creating the potential for both accidental and intentional disruption.

The technical challenge extends to inventory management. If thousands of agents simultaneously add the same limited-stock item to carts, retailers face questions about allocation, fairness, and how to prevent automated systems from dominating access to high-demand products. Flash sales and limited releases, which retailers use to generate excitement and urgency, become difficult to manage when bots can react faster than any human.

Fraud detection systems trained to identify suspicious patterns in human behaviour may struggle with agents that exhibit different browsing and purchasing patterns. A sudden spike in activity from a single user account, or unusual navigation paths through a site, might trigger security flags even when the agent is operating as intended. Retailers must recalibrate their defences without knowing how prevalent agentic commerce will become.

Adoption Remains Marginal

Despite the enthusiasm from technology companies, actual usage of AI agents for commerce remains negligible. Uber reported fulfilling only four rides via third-party agents during the third quarter and approximately 500 Uber Eats orders during the week of Halloween. DoorDash said it completed only several hundred orders placed through external agents, compared with 993 million total orders in its most recent quarter.

Those figures suggest that agentic commerce is far from mainstream adoption. The gap between the capabilities being demonstrated at technology conferences and actual consumer behaviour is substantial. Whether that reflects a temporary lag as the technology matures or a more fundamental mismatch between what developers are building and what shoppers actually want remains unclear.

The low adoption may also reflect practical limitations. Configuring an agent to understand preferences, navigate different retailers, and complete transactions requires a level of technical sophistication that most consumers do not possess. Until agent platforms become as simple to use as conventional shopping apps, they are unlikely to achieve broad uptake.

Regional Implications

The agent access question has particular resonance across Asia, where e-commerce platforms have evolved differently than in Western markets. Super-apps like Grab and Gojek integrate shopping, food delivery, payments, and services within a single ecosystem, creating walled gardens that external agents would struggle to penetrate. In China, platforms such as Taobao and JD.com have built recommendation engines and live-streaming commerce features that assume direct platform control over the shopping experience.

Shopify's decision to open its catalogue may prove more significant for merchants in Southeast Asia and India, where the platform has gained traction as an alternative to dominant local marketplaces. For sellers seeking to reach customers across multiple channels, agent-enabled discovery could provide a distribution advantage. However, the effectiveness depends on whether consumers in those markets adopt agent-based shopping, a question that remains unanswered.

Japanese e-commerce, dominated by platforms like Rakuten and Mercari, has historically prioritised trust and buyer protection mechanisms that involve human verification at key transaction points. Introducing agents into that environment would require retailers to rethink authentication and fraud prevention systems built around the assumption of human users.

The Standards Vacuum

The absence of industry standards for agent access leaves each retailer to develop its own policies and technical controls. That fragmentation may slow the development of the agent ecosystem overall, as developers must navigate a patchwork of restrictions rather than building to a common protocol.

Some technology executives have called for collaborative approaches that would allow agents to access retail platforms under defined terms, potentially including revenue-sharing arrangements or data-use agreements. eBay's statement about ongoing discussions with potential partners suggests the company is open to such arrangements, though no formal framework has emerged.

The alternative is a prolonged period of friction, with agents probing platform defences and retailers responding with increasingly sophisticated blocking mechanisms. That arms race benefits neither side and may ultimately limit the potential of agentic commerce to deliver value to consumers.

For now, the gap between the agent revolution promised by Silicon Valley and the reality of retail platform policies continues to widen. Whether that gap will close through negotiation, standardisation, or continued restriction will shape the trajectory of automated shopping over the coming years.