Alibaba-linked operators allegedly generated over 28.8 million interactions with Anthropic's Claude in just six weeks, marking the largest AI distillation campaign ever recorded
If you thought IP theft in tech was limited to stolen code repos and midnight server raids, welcome to the AI era’s version: tens of thousands of fake accounts systematically milking frontier models until they’ve extracted enough to build a competitive clone. OpenAI and Anthropic are now publicly sounding the alarm that Chinese AI labs are doing exactly that, at industrial scale. Anthropic disclosed that operators linked to Alibaba’s Qwen AI lab ran what it described as the largest distillation campaign ever measured, generating more than 28.8 million interactions with its Claude system between April 22 and June 5, 2026. The operation used roughly 25,000 fraudulent accounts.
The distillation playbook #
Here’s how this works in plain English. “Distillation” means feeding a powerful AI model carefully crafted prompts, collecting its outputs, and then using that data to train a cheaper, smaller model that mimics the original’s capabilities.
This isn’t a one-off incident. Back in February 2026, Anthropic identified a separate wave involving around 24,000 fake accounts that produced 16 million Claude interactions. Those campaigns were attributed to operators connected to DeepSeek, Moonshot AI, and MiniMax.
OpenAI has made similar accusations. The company sent a memo to Congress in February 2026 accusing DeepSeek of systematic distillation of its models.
The Chinese labs reportedly bypass access restrictions by routing through proxy services that manage large networks of fraudulent accounts.
The industry response and the Frontier Model Forum #
In April 2026, OpenAI, Anthropic, and Google formed the Frontier Model Forum, a coalition specifically designed to share intelligence and coordinate defenses against adversarial distillation campaigns.
The technical challenge is genuinely hard. These companies need to distinguish between a legitimate power user making thousands of API calls for a valid enterprise application and a coordinated bot network designed to systematically probe and extract model capabilities. Rate limiting helps, but sophisticated operators can distribute queries across enough accounts to stay below detection thresholds.
Why crypto and AI investors should care #
For the crypto-AI intersection specifically, decentralized AI projects that rely on open-source model architectures may find themselves caught in the crossfire of tightening US export controls. If policymakers respond to these distillation campaigns by imposing stricter restrictions on AI model access, that regulatory overshoot could affect projects building on open infrastructure, not just the bad actors. US policymakers are already weighing additional export controls on AI technologies, and these revelations give them fresh ammunition. Any new restrictions would reshape the competitive landscape for AI chip makers, cloud providers, and the growing ecosystem of crypto projects that depend on affordable access to compute and model infrastructure.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our