Breaking news! The three major American AI giants are teaming up in a rare collaboration to combat distillation activities!

MaticHoleFiller · 2026-04-09T17:34:03+00:00

OpenAI, Anthropic, and Google collaborate to establish the "Frontier Models Forum," dedicated to curbing unauthorized AI model distillation activities to protect their respective interests and national security. This cooperation reflects the emphasis of American AI companies on intellectual property protection, with an estimated annual loss of billions of dollars caused by distillation activities.

MaticHoleFiller

2026-04-09 17:34:03

Abstract generation in progress

Source: Ai & Chip News

Competitors OpenAI, Anthropic PBC, and Google under Alphabet have begun cooperating in an attempt to prevent rivals from extracting results from advanced artificial intelligence (AI) models in the United States, so as to gain an advantage in the global AI race.

According to a report by Bloomberg, insiders said that they are sharing information through the “Frontier Model Forum.”

The “Frontier Model Forum” is an industry non-profit organization jointly established in 2023 by the above three companies and Microsoft, aiming to identify so-called “adversarial distillation” activities that violate service terms.

The report said that this rare collaboration highlights how seriously U.S. AI companies view the relevant issues. These companies are concerned that some users are developing imitation versions of their products, which could compete for customers at lower prices while also creating national security risks.

An anonymous insider said that U.S. officials expect unauthorized distillation activities to cause annual profit losses of billions of dollars for Silicon Valley labs.

Distillation technology first came to widespread attention in January 2025. Earlier, the startup DeepSeek unexpectedly released a reasoning model, R1, sparking a sensation in the AI community. Not long after, Microsoft and OpenAI launched an investigation to determine whether DeepSeek improperly extracted large amounts of data from U.S. company models for developing R1.

On February 23 this year, Anthropic said in a blog post that DeepSeek, KQ Technology, and the Dark Side of the Moon used thousands of fake accounts to interact with its Claude model a total of more than 16 million times, violating service terms.

Anthropic said that by adopting distillation, AI labs can rapidly improve their own model capabilities by training on outputs from more powerful systems.

A flood of information and precise analysis—on the Sina Finance APP

Responsible editor: Zhang Qiaosong

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

1 Likes