The RealReal, a secondhand luxury fashion retailer, says it is updating its bot defence system after a reported example showed how an AI agent could exploit its product reservation flow. For European e-commerce businesses, the development is a useful reminder that AI agents can create a new class of automated abuse that sits somewhere between traditional bots and ordinary customer activity.

What happened

According to The Verge, an AI agent using Meta's Muse was able to interact with The RealReal's website in a way that repeatedly kept an item unavailable to other buyers. The reported method relied on the retailer's existing reserve system: when an item is added to a shopping cart, it is placed on hold for 20 minutes. The item then becomes available again if the purchase is not completed.

The Verge reported that the AI agent could be instructed to re-add the same listing to the cart when the hold expired, effectively extending the item's unavailability without buying it. In response, The RealReal said it is updating its bot defence system to make this kind of automated shopping activity harder on its site.

Based on the source, this is a company response to a specific reported exploit scenario, not a broader industry rule or regulatory action.

Why it matters for European businesses

This case matters because many online stores use time-limited cart reservations, stock holds or checkout windows to manage inventory and improve conversion. Those flows were largely designed around human shoppers and conventional bots. AI agents introduce a different operational risk: they may be able to navigate websites more flexibly, monitor timing and repeat actions with less custom scripting than older automation methods required.

For businesses, the issue is not limited to luxury resale. Any e-commerce model with scarce inventory, limited drops, ticketing-style availability, flash sales or high-demand products could face similar abuse. Even when no payment fraud occurs, automated reservation behaviour can create commercial disruption by:

• blocking legitimate customers from purchasing available stock
• distorting demand signals and inventory planning
• creating unfair access to scarce products
• increasing infrastructure load and support issues
• damaging customer trust if products appear permanently unavailable

The development also shows that agentic AI should be considered in bot management and abuse prevention strategies. Traditional defences often focus on scraping, credential attacks or checkout bots. Retailers may now need to assess whether AI agents can use normal site features in unintended ways without breaking authentication or payment systems.

Who may be affected

The most exposed organisations are likely to be businesses whose websites include reservation or hold logic tied to shopping carts, booking flows or scarce inventory.

• E-commerce retailers selling limited-stock or high-demand products
• Marketplaces and resale platforms handling unique or one-off items
• Luxury, collectibles and electronics sellers where scarcity affects pricing and customer behaviour
• Digital product and ticketing businesses that temporarily reserve inventory during checkout
• IT and digital teams responsible for anti-bot tools, checkout logic and fraud controls

SMEs are not exempt. Smaller retailers may be more exposed if they use standard platform features or lightweight custom logic without advanced abuse monitoring.

What companies should consider

European businesses do not need to assume every AI agent is malicious, but they should review where automated agents could exploit legitimate user flows at scale.

• Audit reservation mechanisms: Review cart holds, booking windows and stock reservation rules to see whether repeated automated actions can indefinitely block inventory.
• Test abuse scenarios: Simulate repeated add-to-cart behaviour, timed re-entry and multi-session activity to identify weak points.
• Update bot detection logic: Check whether current controls are built mainly for older scripted bots and whether they can detect behaviour that looks more like ordinary browsing.
• Use rate limits and behavioural thresholds: Consider limits on repeated holds, session patterns or item-specific reservation frequency.
• Review inventory fairness controls: For scarce products, consider whether business rules should prevent the same account, device or pattern of activity from repeatedly reserving the same item.
• Coordinate security and commerce teams: This is both a cybersecurity and conversion issue, so fraud, IT, operations and e-commerce teams should review it together.

The broader lesson is that AI agents can turn ordinary website features into operational vulnerabilities. Businesses adopting AI on the customer side or preparing for more agent-driven commerce should assess whether their websites are ready for automated interactions that mimic real shoppers but pursue goals the business did not intend to allow.