The RealReal upgrades bot defenses to stop AI agents hoarding goods

What changed at The RealReal?

The luxury consignment platform announced an update to its bot‑defense system that makes it substantially harder for automated agents to log in, add items to a cart, and complete checkout. The move follows a demonstration that an AI assistant could repeatedly re‑add the same listings to a cart, keeping them unavailable to other shoppers for extended periods. According to a report by The Verge (https://www.theverge.com/news/1018713/meta-muse-agentic-shopping-ai-the-realreal), the retailer’s chief technology officer, Luke Friang, confirmed the changes in an email after the story was published.

How did AI agents exploit the reserve system?

The RealReal’s marketplace places a 20‑minute hold on any item that a user adds to their cart. An agent built on Meta’s Muse platform was instructed to monitor the countdown and, the moment the hold expired, immediately re‑add the item to the cart again. By looping this behavior, the agent could keep a product locked for hours, effectively preventing human customers from purchasing it. The Verge’s testing showed that the technique worked reliably across multiple listings, turning a simple reservation feature into a tool for inventory monopolization.

Why does this matter for online retail?

When a non‑human actor can dominate scarce inventory, the shopping experience for real customers deteriorates. Luxury resale sites rely on trust and fairness; a perception that bots can “buy out” coveted pieces threatens brand reputation and could drive shoppers to competitors. Larger platforms such as Amazon have already begun blocking AI‑driven purchases for similar reasons, fearing loss of advertising revenue and erosion of direct customer relationships. The RealReal’s response signals that even niche e‑commerce players must anticipate and mitigate agent‑level abuse as personal‑assistant AI becomes more widespread.

What does the update actually do?

Friang explained that the new configuration raises the friction for any automated login attempt, adds stricter checks when items are added to a cart, and tightens checkout validation. While the system still allows agents to surface product pages and direct users there, the core actions that enable hoarding—continuous cart manipulation—are now blocked. The retailer said it will keep refining these defenses as AI‑powered shopping tools evolve, suggesting a rolling arms race between bot developers and anti‑bot technologies.

How can creators, developers, and researchers explore this scenario with Grafy?

Grafy’s Operator app lets users build and test browsing agents that can interact with live websites under controlled conditions. By linking an Operator workflow to The RealReal’s public product pages, developers can experiment with credential handling, rate‑limiting, and anti‑bot challenge responses without violating the retailer’s terms of service. For example, you could:

1. Use Operator to script a simple “view‑only” bot that fetches item details and returns them to a Grafy Chat session for analysis.
2. Add a step that respects the 20‑minute hold logic, then observe how the updated defenses react.
3. Compare the results with a baseline run on a competitor site that has not yet implemented the new safeguards.

These experiments help AI researchers understand the practical limits of agentic shopping and can inform the design of more ethical assistants. If you are more interested in the policy side, the Reader app can ingest the Verge article PDF (or any related whitepaper) and let you query the text for specifics about the exploit, the retailer’s response, and broader industry trends.

What broader trends does this highlight for AI‑driven commerce?

The incident underscores three emerging patterns:

• Agent‑first shopping – Companies like Meta are marketing AI assistants as the primary interface for purchasing, with average transaction values reported above $200. As these agents become mainstream, retailers must anticipate that bots will act on behalf of users at scale.
• Defensive escalation – Bot‑defense technologies, traditionally focused on scraping and ticket scalping, are now being adapted to stop agents from abusing reservation mechanisms. Expect more sophisticated CAPTCHAs, behavioral analytics, and server‑side throttling.
• Regulatory attention – As AI agents gain the ability to influence market dynamics, regulators may scrutinize practices that give bots an unfair advantage, especially in high‑value secondary markets like luxury fashion.

Where can you keep an eye on future changes?

Grafy’s Watch app can monitor The RealReal’s public pages for updates to terms of service, changes in checkout flow, or new anti‑bot notices. By setting up a watch on the site’s checkout URL, you’ll receive AI‑summarized alerts whenever the HTML or JavaScript changes, giving you an early signal that the retailer has rolled out additional defenses.

Frequently asked questions

How does Meta’s Muse differ from other shopping assistants?

Muse is marketed as an always‑on personal assistant that can schedule events, draft messages, and make purchases. The Verge reported that it is the most common use case for Muse, with average spend over $200 per transaction. Unlike simple voice assistants, Muse can execute multi‑step workflows, such as continuously monitoring a product’s availability and re‑adding it to a cart, which makes it powerful—and potentially disruptive—for e‑commerce platforms.

Will the new bot defense affect legitimate third‑party tools?

The RealReal’s statement indicates that the updated system targets automated login attempts and rapid cart actions, which are typical of bots. Legitimate tools that respect human‑like interaction rates—such as price‑trackers that check a page once every few minutes—should still function. However, developers may need to adjust request pacing and incorporate human‑behavior mimicking to avoid being flagged.

Can I use Grafy to build a compliant shopping assistant?

Yes. By leveraging Operator for web navigation and pairing it with Chat for natural‑language handling, you can design an assistant that asks for explicit user confirmation before each cart addition, respects hold timers, and logs all actions. This approach aligns with the retailer’s expectations for human‑driven interactions while still providing AI‑enhanced convenience.

---

Source: The Verge — AI — The RealReal won’t let AI agents hog designer clothing