OpenAI, Anthropic, and Google Join Forces to Counter Chinese AI Copying

OpenAI, Anthropic, and Google have started collaborating to prevent Chinese competitors from extracting data from their most advanced AI models. The initiative was revealed by Bloomberg on Monday, April 6, 2026, and utilizes the Frontier Model Forum, the nonprofit group the three companies created with Microsoft in 2023.

In practice, they are exchanging information to identify attempts of "adversarial distillation." This technique involves bombarding models with thousands of queries to capture responses, then training new systems at much lower costs, violating terms of use. This move is rare as companies typically compete fiercely in the same market.

OpenAI confirmed to Bloomberg its participation in data sharing on the issue. The company also recalled a memo sent to the US Congress in February this year, accusing the Chinese startup DeepSeek of attempting to "piggyback" on capabilities developed by American labs.

Google and Anthropic declined to comment, as did the Frontier Model Forum itself. However, the report indicates a collective concern: American labs see distillation as a direct threat to the billion-dollar investment they make to maintain technological leadership.

How Distillation Works in Real Life

In short, someone creates fake accounts or uses proxy services to generate millions of interactions with ChatGPT, Claude, or Gemini. Then, they use this data to train a smaller, cheaper model. The result is a competitor that delivers similar performance without spending the same time or computational resources.

So far, the collaboration focuses on detecting extraction patterns and quickly blocking suspicious accounts. There is no indication that companies are sharing code or model architecture—only attack signals.

The next concrete step is already underway: OpenAI indicated that it will continue providing evidence to the US government on the subject, while the group as a whole enhances detection mechanisms within the Frontier Model Forum. With this, the expectation is that barriers against extraction will become more robust in the coming months.