OpenAI and Anthropic are sounding the alarm on industrial-scale intellectual property theft in the AI sector. Chinese AI labs are allegedly using tens of thousands of fraudulent accounts to systematically extract data from frontier models to build competitive clones.
Anthropic revealed the largest "distillation" campaign it has ever measured. Operators linked to Alibaba's Qwen AI lab generated over 28.8 million interactions with Anthropic's Claude system between April and June 2026, using approximately 25,000 fake accounts.
The technique involves feeding a powerful model carefully crafted prompts, collecting outputs, and using that data to train a smaller, cheaper model that mimics the original. This is not an isolated incident. Anthropic identified a similar wave in February 2026 involving around 24,000 fake accounts, attributed to operators connected to DeepSeek, Moonshot AI, and MiniMax.
OpenAI has also accused DeepSeek of systematic distillation in a memo to Congress. The labs reportedly bypass access restrictions by routing through proxy services that manage large networks of fraudulent accounts.
In response, OpenAI, Anthropic, and Google formed the Frontier Model Forum in April 2026. The coalition is designed to share intelligence and coordinate defenses against these adversarial campaigns.
For investors, especially in the crypto-AI space, the implications are significant. Decentralized AI projects relying on open-source architectures could be impacted if US policymakers respond with stricter export controls. Any new restrictions would reshape the competitive landscape for AI chip makers, cloud providers, and projects dependent on affordable compute access.