Skip to content
Case report Open access

Discriminating Against the Tool: AI Shopping Agents, Robot Shoppers, and the Title III Duty to Modify

Sep 2026 · 0 citations · 4 references

TL;DR

The Article carries the argument from the browser to the sidewalk, where embodied robots will soon shop for people who cannot, and shows that the Department of Justice's refusal to let public accommodations ban classes of new mobility technology supplies the method.

Abstract

In September 2026, Amazon began refusing Meta's Muse, a consumer AI agent that shops on a user's instructions, with a pop-up telling customers that continued access by an unauthorized AI agent violates Amazon's Conditions of Use. The fight is usually framed as a contest between platforms over who owns the customer. This Article reframes it around a customer the contest leaves out: the disabled person for whom the agent is the way in. A blind shopper, a person with a mobility impairment, or a person with agoraphobia who cannot enter a store may accomplish the purchase only by delegating it to a tool. When the store bans the tool, it bans that person, and Title III of the Americans with Disabilities Act already has a name for that act. The Article makes three doctrinal moves. First, coverage: in every circuit, including those that confine Title III to physical places, a retailer that runs hundreds of physical grocery stores and uses its website and app to move their goods falls within the nexus doctrine. Second, liability: a blanket agent ban is a facially neutral policy that burdens disabled customers differently and more severely, so Title III's command to modify policies, practices, and procedures applies, and Amazon's own Agent Terms, which let it refuse any agent at its sole discretion, are an eligibility criterion that screens out. The Ninth Circuit's August 2026 holding that a shopping agent is a tool, and that the user is the one who accesses the site, removes the premise that the agent is a separate stranger the store may refuse. Third, defenses: the direct threat defense reaches only threats to others, and a customer's chosen risk to her own account falls outside it; security concerns can be met by conditions such as agent identification, which Amazon's own terms already require; lost advertising revenue is not a change in the nature of selling goods; and a store's offer of its own agent answers the claim only where it delivers a like experience. The Article then carries the argument from the browser to the sidewalk, where embodied robots will soon shop for people who cannot, and shows that the Department of Justice's refusal to let public accommodations ban classes of new mobility technology supplies the method. The Article names the practice at stake assistive delegation, and argues that Title III protects it.

Read PDF

Similar papers

Evaluating Open-Weight E-Commerce Agents with Environment-Grounded Verification

A deterministic and reproducible e-commerce environment is built that precommits each trial's customer and trajectory parameters, including the persona, difficulty, target cart, and an item reveal schedule, and the resulting capability profiles distinguish under-action, over-purchase, unsupported product attributes, an...

Nimit Shah, Haitz Sáez de Ocáriz Borde · 0 citations
Preprint Oct 2026

Preference Is Not Permission: Personalizing the Auto-Purchase Line in Agentic Commerce

Personalized shopping benchmarks ask which item fits the user; we ask how far an agent may commit on that item, and pose learning the individual's auto-purchase line as a task. The boundary is evidence-conditioned and three-valued: whether committing is appropriate depends on the sufficiency of product information (war...

Ryotaro Shimizu, Sai Htaungkham, Mo Fan · 0 citations
#artificial intelligence Preprint Oct 2026

PersonaCUABench: Benchmarking Computer-Use Agents Serving Diverse Personas

Computer-use agents (CUAs) are commonly evaluated against fixed task specifications, but users may omit requirements, add constraints, or require approval before consequential actions. Completing the benchmark instruction does not guarantee serving the person who asked. We introduce PersonaCUABench, which holds 74 task...

Hao-Ming Li, De-Fu Cao, Xu-Ming Hu et al. · 2 citations
Review Oct 2026

When"No Worries"Went Wrong: HRI Lessons for Foundation-Model Robots

When a person declined our robot's invitation to play, Alice turned her shoulder away and said"No worries, it's all good."Only after a brief pause did she drive off. The words and gestures were plausible but their order changed the message. In another episode, Alice informed a host that his visitor had arrived---while...

Eliahu Khalastchi · 0 citations
#artificial intelligence Preprint Oct 2026

Get In, Decide, Prove: A Merchant-Side Readiness Benchmark for Agentic Commerce

AI agents now browse and buy on people's behalf, and benchmarks measure how well the agents perform. Far less is known about the other side of the transaction: whether a merchant's website lets an agent in, publishes terms precise enough to decide a purchase, and leaves evidence of what was agreed. This paper presents...

Adrian de Valois-Franklin, Alexander Bogdan, Anas Ibrahim et al. · 0 citations

Trust and Task Completion in the World of Consumer AI Agents

An evaluation is built that scores trust and completion on the same runs, in a simulated world of businesses with their own websites, inboxes, and phone lines, and of people who write back, to measure Fo, Wajo's personal assistant, against a base model with basic instructions on three foundation models, and against the...

Jeroen Olieslagers, Eduardo E. P. Pujol, Gal Zahavi et al. · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.