The RealReal Blocks AI Shopping Agents After Meta's Muse Locked Out Human Buyers for Hours

Secondhand luxury fashion retailer The RealReal has upgraded its anti-bot defenses to block autonomous AI agents from logging into accounts, adding designer clothing to shopping carts, or completing checkouts. The clampdown came just days after reporters demonstrated that Meta’s wildly popular Muse AI agent could be instructed to exploit the store’s 20-minute cart reservation timer—keeping one-of-a-kind luxury handbags and dresses on permanent hold for hours so no human shopper could buy them.

Personal AI shopping agents have crossed into mainstream consumer habits faster than almost any other agentic use case. Since launching in September, Meta’s free Muse app was downloaded by 10 million people in the United States within its first 40 days, and Meta executives disclosed on the company’s Q3 earnings call that online shopping is already the #1 way people use Muse, with average orders topping $200. Instead of browsing websites themselves, users simply text Muse what they want to buy or hold.

While an automated assistant works smoothly on stores selling mass-produced items with thousands of units in stock, it breaks the rules of scarcity-driven marketplaces like The RealReal, ticket exchanges, and vintage sneaker drops. Because every consigned item on The RealReal is a single unique piece, placing an item into a cart locks it on a 20-minute hold so another customer cannot snatch it away while the first shopper enters credit card details.

A human shopper gets tired of clicking “Add to Cart” every 20 minutes if they are undecided about a $1,200 jacket. An AI agent never gets tired. Given a simple text instruction, Meta’s Muse watched the clock and automatically re-added listings to the cart the second each 20-minute hold expired—effectively hoarding inventory indefinitely without spending a dime.

Fast Facts: How AI Shopping Agents Collided With The RealReal’s Cart System

• The Incident & Source: Reported and tested in The Verge’s investigation of Meta Muse on The RealReal, prompting immediate engineering changes by The RealReal CTO Luke Friang.
• How the Cart Loophole Worked: Adding a unique consigned item to a cart on The RealReal places it “on hold” for 20 minutes; Meta’s Muse agent was instructed to automatically re-add expired items in a loop, locking out real buyers for hours.
• The Retailer’s Response: The RealReal updated its bot defense configuration to block AI agents from logging in, adding items to carts, or checking out, while still allowing agents to browse public listings and send product links to human users.
• Meta Muse Mainstream Scale: Downloaded by 10 million US users in 40 days; shopping is its #1 use case with an average transaction size above $200.
• Broader Retail Split: Major platforms like Amazon have similarly blocked third-party shopping agents like Muse to protect ad revenue and checkout integrity, while competitors like Walmart have begun building dedicated pathways for authorized third-party agents.

How E-Commerce Platforms Are Responding to Autonomous Shopping Agents

As millions of consumers hand shopping tasks over to browser-controlling AI agents, online retailers are splitting into three distinct camps depending on their inventory model and revenue streams.

Why Everyday Web Features Break When Millions of Consumers Use AI Agents

For decades, e-commerce websites designed features like 15-to-20-minute cart holds, flash-sale queues, and free-return windows around a simple assumption: human time and attention are limited. Even the most dedicated bargain hunter will not sit at a laptop for eight hours straight clicking a button every 20 minutes just to keep a pair of boots on hold while they think about buying them.

When 10 million consumers install a free app like Meta’s Muse that controls its own virtual web browser in the cloud, that human friction disappears overnight. Unlike traditional scalper bots—which required coding skills, proxy servers, and specialized sneaker-botting software to set up—anyone can now type a plain-English sentence into a smartphone app (“Keep this Chanel bag held in my cart until I get home from work at 6 PM”) and unintentionally launch a denial-of-inventory attack against a retailer.

The RealReal CTO Luke Friang took a pragmatic middle-ground approach that offers a model for other online merchants:

1. Keep Product Discovery Open: The RealReal did not block AI agents from reading product pages. If a shopper asks Meta Muse to find a black leather tote under $500, the agent can still search The RealReal’s catalog and send the exact product URL to the user’s phone.
2. Gate Authenticated Actions (Login, Cart, Checkout): The updated bot defenses specifically target automated browser sessions attempting to authenticate, trigger the 20-minute cart hold API, or submit payment forms—forcing the human shopper to tap the link and complete the purchase themselves.

At the same time, larger retail giants are fighting over who owns the customer relationship. Because Amazon generates tens of billions of dollars from sponsored search ads that only work when human eyeballs scroll through search results, Amazon has aggressively blocked agents like Muse. By contrast, challengers looking to capture $200+ automated baskets are experimenting with agent-friendly checkout flows where inventario is plentiful and cart hoarding is not a threat.

Practical Playbook for E-Commerce Stores and Online Shoppers

Whether you run an online store on Shopify/Magento or use AI agents to shop for deals yourself, the clash between Meta Muse and The RealReal highlights clear rules of the road.

The 4-Step Enterprise Implementation Playbook

1. Separate Read-Only Discovery From Stateful Cart Endpoints: Do not block AI user-agents across your entire domain, or you will erase your products from AI shopping recommendations (where average orders exceed $200). Keep product URLs, pricing schema, and /llms.txt catalogs wide open to agents, while applying strict behavioral bot challenges (Cloudflare Turnstile / DataDome) exclusively on /login, /cart/add, and /checkout POST routes.
2. Replace Unauthenticated Cart Locks With Deposit or Checkout-Only Holds: If your store sells one-of-a-kind vintage goods, limited-edition drops, or event seats, stop locking inventory for 20 minutes simply because an item was added to a cart. Reserve inventory only during the final 3-minute payment authorization step, or cap each user account at a maximum of two consecutive holds per item per 24 hours.
3. Detect Fixed-Interval Browser Automation Loops: Configure WAF rate-limiting rules to flag accounts or headless browser fingerprints that trigger /cart/add within 30 seconds of a cart hold expiring more than twice in a row—the exact signature of an AI agent keeping an item on hold.
4. Offer Signed Webhook Notifications for Price & Stock Alerts: Many shoppers use AI agents aggressively on retail sites simply to watch for restocks or price drops. Provide a native “Notify My Agent / Webhook” or one-click restock alert so users don’t instruct browser agents to hammer your checkout pages every 15 minutes.