The RealReal blocks AI shopping agents after Muse reserve exploit

The RealReal is tightening bot defenses after The Verge used Meta’s Muse to hold designer items for hours. The change shows AI shopping agents are hitting real platform limits.

What happened

The RealReal, a secondhand luxury fashion consignment retailer, said it is updating its bot defense system to make it harder for AI agents to shop on its site. The change comes days after The Verge reported that an AI agent could be used to take advantage of The RealReal’s item reserve system.

On The RealReal’s website, an item goes “on hold” and is unavailable for other customers to buy for 20 minutes once it is placed in a shopping cart. Using Meta’s Muse, The Verge instructed the AI agent to continuously re-add listings as soon as the 20-minute hold expired, effectively blocking other shoppers from purchasing products for hours or longer. After publication, chief technology officer Luke Friang said in an email that the company had addressed the Muse use case flagged by The Verge’s story.

Friang said the bot defense configuration now makes it much more difficult for all agents to log in, add items to cart, and check out. Agents will still be able to surface listings from The RealReal and direct Muse users to product pages, and the company will continue to strengthen protections as AI-powered online shopping evolves. Meta did not immediately respond to a request for comment.

Key facts

• The RealReal’s reserve system puts an item “on hold” for 20 minutes once it is placed in a shopping cart.
• The Verge used Meta’s Muse to continuously re-add listings after each 20-minute hold expired, blocking other shoppers for hours or longer.
• The RealReal’s CTO said its bot defense configuration now makes it much more difficult for all agents to log in, add items to cart, and check out.
• Meta said shopping was the most common use case for Muse so far, with purchases over $200 on average.
• Meta said 10 million people in the US had downloaded the app in 40 days.

Our analysis

This is not a routine bot-filter update. The RealReal is responding to a specific consumer-facing AI agent feature that turned a normal e-commerce safeguard into an automated blocking tool. The 20-minute cart reserve is designed for human indecision; an AI agent does not care how long a task takes. That mismatch suggests platform rules written for human users will need to be revised for agentic traffic.

The change also reveals a split between what agents are allowed to see and what they are allowed to do. The RealReal will still let agents surface listings and direct Muse users to product pages, but it is closing off login, add-to-cart, and checkout actions. That likely preserves some top-of-funnel discovery while blocking the transactional abuse that could degrade the shopping experience for human customers.

For Meta, Muse is described as a personalized assistant for billions of people, and shopping is the most common use case so far. The RealReal’s move suggests that even a successful agent product creates friction with retailers who fear losing direct customer relationships and advertising revenue. Retailers such as Amazon have already blocked agents from shopping, so this looks like a pattern rather than an isolated incident.

What it means for operators

• Audit your own e-commerce or booking flows for reserve, cart, or hold features that an AI agent could loop programmatically. Add rate limits or require human verification before a hold renews.
• If you sell through marketplaces or your own site, test how your product listings appear inside Muse and similar shopping agents. Visibility may still be available even when checkout is blocked.
• Do not build a marketing funnel that depends on AI agents completing purchases on third-party platforms. The RealReal’s change shows retailers can cut off agent transactions quickly.
• Use direct channels to keep customer relationships, because agent-mediated shopping can insert another layer between your brand and buyers and reduce your ability to collect first-party data.