Three companies have fought most fiercely in the AI industry until now. OpenAI, Anthropic, and Google, they pour billions into stealing each other's customers, poaching each other's talent, and crushing each other's model benchmarks. Yet on April 6, 2026, these three companies sat at the same table and began sharing intelligence. What made enemies join hands was a single threat: China.
What Actually Happened: 24,000 Fake Accounts and 16 Million Conversations
The figures Anthropic disclosed are startling. DeepSeek, Moonshot AI, and MiniMax, three Chinese AI startups, conducted more than 16 million conversations with Claude through roughly 24,000 fraudulent accounts. The purpose was not simple. So-called "adversarial distillation," a technique of collecting a frontier model's response patterns at scale to train one's own model. It is a way to replicate frontier AI capability at far lower cost. The US government estimates that through this technique American AI companies lose billions of dollars' worth of competitive advantage each year.
Why This Is More Serious Than People Think
Adversarial distillation is not simple IP theft. It is a strategy that fundamentally shakes the balance of the AI supremacy race. While OpenAI invests years into its $500 billion Stargate project and Anthropic raises $40 billion from Google to build out compute infrastructure, Chinese competitors can replicate the result at far lower prices and erode the global market. With Anthropic's Claude 3.5 Sonnet holding 32% of the enterprise LLM API market and OpenAI's GPT-4o following at 25%, if that market share collapses to distilled replicas, the entire revenue model of Western AI companies wobbles.
Hidden Insight: A Matter of Survival, Not Competition
The real meaning of this alliance is not the cooperation itself but the scale of the threat that forced it. OpenAI, Anthropic, and Google co-founded the Frontier Model Forum with Microsoft in 2023, but actual intelligence sharing has been extremely limited until now. This move is a declaration that they have crossed a threshold where individual defense can no longer cope. There is a historical precedent: just as GM and Ford cooperated on fuel-economy technology standardization in the face of the rise of Japan's auto industry in the 1970s, an external threat turns competitors into allies. This structure matters for Korean companies too, because the moment Kakao, Naver, and SKT must choose American or Chinese AI, the existence of this alliance could become the criterion for which side is the more trustworthy supplier. The bear case, however, is that critics argue distillation is hard to prove and merely a terms-of-service violation rather than a crime, that determined actors will simply route through intermediaries, and the risk is that an alliance of rivals could tip into anticompetitive coordination that regulators eventually punish.