AI Buying Agents Raise New Accountability Questions

Earlier this year, eBay updated its user agreement to explicitly ban third-party “buy for me” agents and AI chatbots from interacting with its platform without permission. The move highlights a broader question facing online marketplaces as AI shopping agents become more capable: Who controls the transaction when software acts on the buyer’s behalf?

Blaine Nielsen, president of retailers at Rithum, suggested that eBay’s action makes an important statement that autonomous purchasing should not operate without accountability.

Rithum, formed by the 2023 merger of CommerceHub and ChannelAdvisor, is an end-to-end commerce operations platform connecting brands, retailers, and suppliers to automate product listings, manage dropship programs, and scale fulfillment across hundreds of online marketplaces.

He expects a human-in-the-loop partnership to always exist. But as marketplaces build technical standards and AI earns more trust, the concept will likely evolve. Consumers will approve spending thresholds, delivery preferences, and trusted-seller rules, with AI handling execution within those limits.

“Setting clear boundaries gives marketplaces the ability to protect themselves in the short term while establishing guardrails, formal access points, and monetization models for AI-driven transactions as they continue to mature,” he told the E-Commerce Times.

Why AI Buying Agents Aren’t Ready Yet

Nielsen noted that AI buying agents cannot yet fully deliver confidence and control in the online shopping experience. Rithum’s recent survey found that only 15% of shoppers have used AI to complete a purchase.

Consumers want less friction. However, today’s AI buying agents cannot yet deliver that convenience without introducing new risks and challenges, he observed. Detecting highly sophisticated, large language model (LLM)-driven AI agents that mimic human browsing behavior is notoriously difficult.

eBay’s blocking of AI agents risks a long-term consumer backlash in favor of platforms that actively welcome agentic shopping, he warned. They are more likely to build long-term consumer trust if they balance automation with proper safeguards and human oversight.

“Platforms that actively welcome agentic shopping may reap early benefits of added simplicity to the shopping experience, but it’s more likely they run into obstacles, like upset customers over unauthorized purchases or incorrect items, in the future,” Nielsen said.

The Limits of Bot Detection

Sophisticated AI agents are becoming increasingly difficult to identify because they can mimic legitimate human browsing behavior, including realistic navigation patterns, residential IP usage, and interactions with accessibility tools.

Marketplaces can limit abusive automation through behavioral analytics, device intelligence, reputation scoring, and risk-based verification. Yet every detection system faces the same balancing act: stopping bad actors without creating friction for legitimate users.

“The larger challenge is avoiding friction for legitimate users, including high-value shoppers, enterprise buyers, and customers using assistive technologies such as screen readers or voice navigation,” said Nielsen.

As AI agents become more advanced, he sees marketplaces moving away from distinguishing between humans and bots toward focusing on whether activity is authorized, trustworthy, and policy-compliant.

“In practice, enforcement will increasingly rely on identity, account reputation, permissions, and rate limits rather than traditional bot-detection methods alone,” he suggested.

Who’s at Fault When Payment Issues Occur

If an autonomous AI agent purchases a counterfeit item, enters the wrong shipping address, or misinterprets a listing, determining legal and operational responsibility may remain a gray area.

According to Nielsen, legal and operational responsibility generally remains with the human user or business deploying the AI agent. AI systems are not independent legal actors.

If an AI agent acts outside its authorized parameters, responsibility will usually depend on where the failure occurred. The fault could stem from user instructions, the seller’s listing, marketplace controls, or the AI provider’s system behavior.

From an e-commerce perspective, AI-agent transactions will likely be treated similarly to other delegated purchasing systems, Nielsen offered. The account holder remains responsible. Marketplaces and AI providers must maintain strong authentication, audit trails, disclosures, and dispute-resolution processes.

“Over time, the industry may evolve toward formal agent authorization frameworks with defined permissions, spending limits, and liability boundaries,” he said.

Question of Intent

Nielsen pointed to a challenge for traditional e-commerce trust models, which have long been built around human intent. Agentic commerce complicates that assumption, particularly when disputes arise.

He explained that shoppers could use AI-assisted purchasing as grounds to seek refunds or return items outside the normal return window by claiming they did not personally make the purchase.

“We saw this friction increase when chatbots and automated call centers emerged. Now there’s a risk AI buying brings this to the next level if there aren’t proper guardrails in place to mitigate the risk,” he warned.

When AI Agents Start Bidding

eBay’s bidding and auction format raises additional legal concerns because timing, bidding psychology, and competitive dynamics make autonomous AI participation in auctions inherently more complex.

“Similar to an unintentional purchase, autonomous agents could overbid, react too quickly based on bidding patterns and timelines, or create artificial price inflation because they can’t balance winning a bid with true shopper value,” Nielsen said.

He cited an even greater risk that the seller and buyer would have to settle a dispute without a human-in-the-loop in bidding and auction formats. Having proper boundaries, such as spending limits and human user purchase confirmation, could help weigh the risks and rewards of using an AI agent to participate in auctions, especially when a shopper is bidding on several items in a short timeframe.

Nielsen views eBay’s decision to ban unauthorized buy-for-me agents as an early indication that auction marketplaces could face dynamics similar to those seen in high-frequency trading. Autonomous bidding creates speed and algorithmic advantages that could make it nearly impossible for humans to participate.

“This reinforces the need for more accountability and intent in agentic shopping experiences to keep human trust at the center of each transaction,” he urged.

How to Handle Future Machine-to-Machine Commerce

While much of today’s debate focuses on how AI agents interact with marketplaces, Nielsen believes the longer-term impact may be the rise of machine-to-machine commerce, where software increasingly represents both buyers and sellers.

According to Nielsen, agentic shopping represents the next step in the broader shift toward more connected, data-driven commerce. Brands and retailers need accurate, real-time product information, fulfillment visibility, pricing consistency, and trusted marketplace connectivity now more than ever.

He advises retailers to prepare for a future in which seller-side AI agents negotiate directly with buyer-side agents. Many retailers and marketplaces are already exploring versions of seller-side AI agents, particularly for dynamic pricing, inventory management, product recommendations, and automated customer engagement.

“Over time, e-commerce likely evolves toward agent-to-agent interactions, where buyer agents and seller agents negotiate around price, fulfillment speed, bundles, warranties, and promotions in real time,” he predicted.

Retailers will want their own intelligent agents to ensure their products, policies, and brand preferences are represented accurately within these automated purchasing ecosystems. As AI-driven commerce evolves, marketplaces will likely establish official APIs, certification programs, and trusted-access frameworks for approved buying agents, he concluded.

“From a platform perspective, unmanaged scraping and autonomous purchasing create operational, fraud, and infrastructure risks, while authenticated AI-agent access creates opportunities for governance, rate controls, identity verification, and monetization,” he said.