OpenAI, Anthropic, and Google Are Teaming Up to Stop China From Copying Their AI — But at What Cost?

The New AI Cold War Just Got Official

On April 6, 2026, something unprecedented happened in the AI industry: OpenAI, Anthropic, and Google — three fierce competitors who normally guard their secrets from each other as jealously as they guard them from anyone else — announced they're working together to stop Chinese AI companies from copying their models.

The collaboration runs through the Frontier Model Forum, an industry nonprofit the three companies founded with Microsoft back in 2023. Until now, the Forum has mostly been a venue for safety pledges and government-friendly PR. This is the first time it's been activated as an actual threat-intelligence operation against a specific external adversary.

What Is "Adversarial Distillation" Anyway?

The technique these companies are worried about is called distillation. Here's how it works in plain English: a rival company creates thousands of accounts on, say, Claude or ChatGPT. They feed those accounts carefully designed prompts — millions of them — and collect all the responses. Then they use those prompt-response pairs to train their own, cheaper AI model that mimics the behavior of the original.

Think of it like a student who doesn't study the textbook but instead copies every answer from the smartest kid in class. Eventually, the student's answer sheet looks pretty similar to the smart kid's — without ever understanding the underlying material.

The Scale of the Problem

The numbers are staggering. In February 2026, Anthropic identified three Chinese AI labs — DeepSeek, Moonshot AI, and MiniMax — that had allegedly created approximately 24,000 fake accounts and run more than 16 million exchanges with Claude. That's an industrial-scale operation, not a casual experiment.

This drew serious scrutiny after January 2025, when DeepSeek released its R1 reasoning model. R1 matched the performance of leading American systems at a fraction of the development cost, sparking widespread suspicion that distillation played a role.

How the Intelligence Sharing Works

The new arrangement is modeled after how cybersecurity firms share threat intelligence. When one company detects a suspicious pattern — say, a cluster of accounts all sending unusually structured prompts — it flags that pattern for the others. The goal is to make it harder for distillation attacks to succeed by ensuring all three companies can recognize and block the same tactics simultaneously.

In practice, this means if DeepSeek gets caught using a particular technique against OpenAI's models, both Anthropic and Google can immediately start watching for the same approach.

Why You Should Care (Even If You're Not in China)

This coalition sounds reasonable on the surface — companies protecting their intellectual property from theft. But there are some real tensions worth thinking about:

Privacy implications. To detect distillation attacks, these companies need to analyze user behavior patterns at scale. They're looking at how people use their tools, what prompts they send, and whether those patterns match known attack signatures. That surveillance infrastructure doesn't just apply to Chinese labs — it applies to everyone.

The openness question. The AI industry has benefited enormously from open research and shared knowledge. This move toward treating outputs as protected intellectual property and monitoring how people interact with models represents a philosophical shift toward a more closed, controlled ecosystem.

Geopolitical escalation. Framing this as "American AI vs. Chinese theft" simplifies a complicated picture. Chinese labs have also produced genuine innovations, and the global AI research community has historically thrived on cross-border collaboration. Drawing hard lines could accelerate a technology decoupling that ultimately slows progress for everyone.

Terms of service as weapons. The coalition is framing distillation as a terms-of-service violation. But ToS agreements were designed to govern individual users, not to serve as the legal framework for an international technology conflict. There's a significant gap between "you can't scrape our API" and "we're conducting coordinated counter-intelligence operations."

The Uncomfortable Questions

If these companies can monitor for distillation patterns, what else can they detect about how you use their tools? If they're sharing intelligence about suspicious users with each other, what happens when a legitimate researcher gets flagged? And who decides what counts as "adversarial" use versus just creative prompting?

The anti-distillation coalition might be a necessary response to real intellectual property theft. But it's also a step toward AI companies acting more like intelligence agencies than tech platforms — and that shift has implications that go far beyond the US-China AI race.