OpenAI, Google, Anthropic join hands to curb AI model copying by Chinese rivals

Leading artificial intelligence companies including OpenAI, Google and Anthropic have begun coordinating efforts to counter what they describe as unauthorised extraction of their model capabilities by overseas competitors.

The companies are sharing intelligence through the Frontier Model Forum, a nonprofit body they co-founded with Microsoft in 2023. The initiative is aimed at identifying and preventing "adversarial distillation" -- a technique where outputs from advanced AI systems are used to train competing models without permission.

The collaboration marks an unusual alignment among rivals, reflecting growing concern that some actors, particularly in China, may be replicating proprietary AI systems at significantly lower cost. Industry estimates suggest such practices could be costing US companies billions of dollars annually in lost revenue.

Also read: OpenAI outlines AI economy vision with 4-day work weeks, robot tax proposals

Distillation itself is a widely used method in AI development, where a smaller "student" model learns from a larger "teacher" model. However, companies argue that unauthorised use of this technique to recreate commercial systems crosses legal and ethical boundaries.

OpenAI has publicly raised concerns about Chinese startup DeepSeek, alleging that it attempted to replicate capabilities developed by US firms. The issue gained prominence following the release of DeepSeek's R1 reasoning model in early 2025, which triggered scrutiny across the industry.

Also read: Google launches offline AI dictation app to rival speech-to-text tools

US AI firms have also warned that such practices could lead to the creation of powerful models without built-in safety safeguards, increasing the risk of misuse. Concerns range from economic competition to broader national security implications.

The rise of open-weight models in China, which make parts of their systems publicly accessible, has added to the pressure on US companies that rely on proprietary models and paid access to recover massive infrastructure investments.