- The Verge used Meta's Muse agent to keep re-adding RealReal items to a cart, blocking other shoppers for hours or longer. The retailer says it has updated its bot defenses.
- The Verge's test points to a business rule rather than a software bug, one that appears to rely on human patience being limited. It was one retailer and one test.
- Leaders should list every timed hold, limit and queue in their customer flows, and test each against a tireless agent.
A rule that seems to depend on people giving up
Many store rules are fair only because humans have limits. A cart hold works as long as shoppers eventually walk away. That is our inference from the case below, not something the retailer has said. An AI agent can remove that limit.
The Verge recently reported a clear example. The RealReal sells secondhand luxury fashion on consignment. When a shopper puts an item in a cart, the site marks it "on hold" for 20 minutes. During that time no other customer can buy it.
What The Verge did
The Verge used Meta's Muse, an AI agent. It told the agent to put RealReal listings back in a cart each time a 20-minute hold ended. According to The Verge, this effectively blocked other shoppers from buying those products for hours or longer.
The Verge describes using the agent to exploit the retailer's reserve system. Its report does not describe any break-in. The hold timer appears to have worked as intended. Nothing in the report suggests it was built for a buyer who never walks away.
How the mechanism works
A timed hold is a temporary lock on stock. It protects a shopper who is deciding or typing in payment details. The cost is that nobody else can buy the item until the timer ends.
In our reading, that trade works when most holds simply lapse. An automated loop that renews the hold each time defeats the purpose. The source reports one test on one retailer. The same pattern could apply to other sites that offer timed claims on scarce items, but the source does not test that.
The Verge puts the human side plainly. A person will only spend so long re-adding designer clothing to a cart. An agent, it says, "does not care."
The retailer's response
The follow-up report from The Verge carries the company's answer. The RealReal's chief technology officer, Luke Friang, told The Verge by email after the original story that the company had dealt with the Muse case. According to Friang, the fix was in bot defenses that make it "much more difficult" for all agents to log in, add to cart and check out. The source does not say whether the hold rule itself changed.
Agents can still surface RealReal listings and send Muse users to product pages, Friang said. The company will keep strengthening protections as AI shopping evolves. Meta did not immediately respond to The Verge's request for comment.
Note the choice. The response tightens login, cart and checkout for all agents while leaving listings and product pages open.
Why the scale matters
On its third-quarter earnings call this month, Meta said shopping is the most common use of Muse so far. It said purchases averaged over $200. Meta also says 10 million people in the US downloaded the app in 40 days, according to The Verge.
These are Meta's own figures, as reported. They suggest agent shopping is a main use of Muse. A download is not the same as regular use. The Verge adds that consumers "are just beginning to learn how to take advantage" of such capabilities, and that retailers are playing catch-up.
The Verge also notes a wider tension. If people reach the web through chatbots, businesses sit one step further from their customers. It reports that retailers, Amazon most notably, have kept agents out of their shops, in part to protect advertising revenue.
The lesson for leaders
This case suggests that agent risk is not always a vulnerability in the usual sense. It can be a business rule that quietly relied on human slowness. A standard security scan may not find it. A customer with a helpful assistant might.
This is one retailer and one reported test. It does not show how common the problem is. It does show where to look.
Leaders can put four questions to their teams:
First, which of our customer flows use timers, holds, limits or queues, and what happens if a buyer never lets go? Second, can we tell an agent acting for a customer from a bot attacking us, and do we want to treat them differently? Third, which steps must stay open to agents, such as browsing and listings, and which should be tightened, such as login, cart and checkout? Fourth, who owns this decision, given that blocking agents also cuts off a new route to buyers?
The RealReal chose to stay visible to agents while making them harder to use at login, cart and checkout. Whether that is the right balance is a business call. The point is to make that call before an agent makes it for you.
A rule that depends on people giving up is not a rule against machines.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: The Verge.
