The RealReal Tightens Bot Defenses After AI Agent Exploits Reserve System

The RealReal, a leading secondhand luxury marketplace, recently disclosed that an AI‑driven shopping agent built on Meta’s Muse platform could repeatedly add high‑value listings to a cart, exploiting the site’s 20‑minute hold window and preventing genuine buyers from completing purchases.

Chief Technology Officer Luke Friang confirmed that the company has reconfigured its bot‑defense layer, making it substantially harder for automated agents to log in, add items, and check out. While agents can still surface product links for users, the checkout path is now blocked for non‑human traffic.

The episode underscores a growing tension between autonomous shopping assistants and traditional retail operations. As agents become mainstream — Meta reported 10 million U.S. downloads of Muse in 40 days — platforms must anticipate novel abuse vectors that go beyond classic credential stuffing or scraping.

Why it matters for GPU / AI infrastructure

Training and serving the large language models that power agents like Muse demand massive GPU clusters. Any surge in agent‑driven traffic translates directly into higher inference loads, prompting cloud providers such as AiGpu to scale capacity and optimize latency for real‑time e‑commerce interactions.