What did the US government announce about AI theft?

The White House Office of Science and Technology Policy said it will combat “industrial-scale” efforts by foreign entities to copy US AI models. The focus is on coordinated campaigns allegedly linked to China. The goal is to protect American AI innovation.

What is meant by “AI model distillation”?

Distillation is a process where one AI model learns from another by analysing its outputs. In this case, it is being done without permission. This allows competitors to replicate capabilities at lower cost.

How are foreign firms allegedly accessing US AI models?

Authorities say attackers use thousands of proxy accounts to avoid detection. They also apply jailbreaking techniques to bypass safeguards. This enables extraction of proprietary model behaviour.

Why is this a concern for the US AI industry?

It allows competitors to build similar models without the same development costs. This undermines the competitive advantage of US firms. It also raises intellectual property and security risks.

Which companies have raised concerns about these attacks?

Anthropic has accused Chinese firms of targeting its Claude models. It reported millions of interactions from tens of thousands of suspected fraudulent accounts. This suggests large-scale coordinated activity.

Which foreign firms were named in the allegations?

Anthropic identified companies including DeepSeek, Moonshot AI, and MiniMax. They were accused of extracting capabilities from US models. The claims have not been independently verified in this context.

How do cost differences highlight the issue?

Advanced US models can cost $5 to $30 per million tokens, while some Chinese models cost as little as $0.26. This gap reflects lower development costs. Distillation may contribute to this pricing advantage.

What risks do “distilled” AI models pose?

These models may lack built-in safety and security controls. This could make them less reliable or more prone to misuse. It raises concerns about global AI standards.

US targets China over AI model theft

Written by Mahathir BayenaPublished Apr. 24 2026Last Updated Apr. 24 2026

The White House has pledged to combat what it describes as “industrial-scale” efforts by Chinese-based entities to copy artificial intelligence models developed by US companies.

Michael J. Kratsios said authorities have evidence that foreign actors are targeting leading American AI firms using proxy accounts and jailbreaking techniques to extract proprietary capabilities.

The administration warned that these campaigns allow competitors to replicate selected performance benchmarks at significantly lower costs while bypassing original development efforts.

“Models developed from surreptitious, unauthorised distillation campaigns like this do not replicate the full performance of the original,”

Kratsios said, adding they still enable cheaper competing products.

The concerns follow allegations by Anthropic that Chinese firms including DeepSeek, Moonshot and MiniMax conducted large-scale distillation attacks involving millions of interactions across tens of thousands of accounts.

Officials said the US will increase coordination with the private sector, share threat intelligence, and explore measures to hold foreign actors accountable as part of a broader defensive strategy.

The issue highlights intensifying global competition in AI development, as pricing disparities between US frontier models and lower-cost alternatives from China continue to widen, raising concerns over security and technological leadership.