The Data Extraction War: Why LLM Supply Chain Risks Now Matter for Retail CX

For retail operators betting their customer experience (CX) on the reasoning capabilities of large language models (LLMs), the latest clash between AI giants isn't just a corporate feud—it's a signal of a volatile supply chain. When the underlying intelligence of a model is compromised or replicated through illicit means, the proprietary advantage a brand builds into its AI-driven service can evaporate.

As first reported by the Financial Post, OpenAI has accused Chinese rival Moonshot AI of orchestrating a wide-scale effort to extract data from GPT systems. OpenAI claims this activity was designed to reproduce the reasoning and capabilities of its most advanced models. In a blog post published Wednesday, OpenAI revealed it detected thousands of attempts by users linked to Moonshot to uncover hidden information regarding how its models solve problems. These coordinated efforts reportedly began in early July and reached a peak of 16,000 requests later that month. While OpenAI noted it could not attribute every operator to one entity, it stated that Moonshot-associated users played a significant role.

This conflict centers on a technique known as "distillation," where outputs from a high-performing model are used to train a less sophisticated one. While the Financial Post notes that distillation is a common industry practice, it becomes a violation of terms of service when executed at an industrial scale. This is a critical point for commerce leaders: the very efficiency that allows a rival to build a chatbot at a fraction of the cost often comes from the systematic extraction of proprietary knowledge from American firms.

This is not an isolated skirmish. The Financial Post reports that OpenAI, Alphabet Inc.’s Google, and Anthropic PBC have been coordinating to counter what they describe as "adversarial forms" of distillation, specifically targeting firms from China. This effort is mirrored by concerns from the Trump administration regarding the systematic extraction of U.S. tech knowledge.

From an operator's perspective, this instability is a risk factor. Moonshot's Kimi K3 model has already challenged assumptions about Chinese AI capabilities and drawn scrutiny in Washington for its rapid progress. As Caroline Zier, lead for strategic national security policy initiatives at OpenAI, told the Financial Post, the goal is to ensure the U.S. remains the leader in the field. While Zier expressed support for an open-weight model ecosystem, she emphasized that the current focus is on stopping violations of terms of service and investing in stronger protections.