The White House has issued a warning, accusing foreign entities, particularly those based in China, of conducting "industrial-scale" operations to copy advanced American AI models. These alleged campaigns reportedly use tens of thousands of fake accounts to extract proprietary knowledge, focusing on the technique of 'distillation.' Major U.S. AI firms, including OpenAI and Anthropic, have previously warned about DeepSeek's attempts to replicate their model capabilities. Furthermore, officials noted that distilled models pose security risks, as they can be stripped of necessary safeguards. In response, the U.S. administration plans to enhance information sharing and develop measures to hold foreign actors accountable for these activities.
Ad slot
The White House has accused foreign entities, primarily from China, of conducting 'industrial-scale' campaigns to replicate advanced American AI models. A memo from the White House Office of Science and Technology Policy detailed these efforts, alleging systematic extraction of proprietary AI capabilities.
Allegations of AI Model Replication
According to Michael Kratsios, White House director of the office of science and technology policy, these campaigns involve sophisticated methods to bypass detection and steal intellectual property.
Methodology: The alleged efforts utilize tens of thousands of surrogate accounts to remain undetected.
Goal: To systematically extract capabilities from leading American AI models, exploiting U.S. innovation.
The Technique: AI Model Distillation
At the core of the accusations is a process known as "distillation." This technique transfers knowledge from a large, powerful AI model into a smaller, more cost-effective version.
Industry Use: Distillation is a common practice for training AI models.
Concerns: Companies like Anthropic and OpenAI have previously warned that this process has been used to unfairly mimic their proprietary model capabilities.
Key Companies Implicated
Ad slot
The Chinese AI startup DeepSeek has been a central focus of these accusations. Reports indicate multiple instances of alleged replication:
OpenAI's Warning: In February, OpenAI alerted lawmakers that DeepSeek was attempting to replicate the performance of ChatGPT and other frontier models using distillation.
Anthropic's Findings: Anthropic also stated in February that it identified "industrial-scale campaigns" involving DeepSeek and two other AI labs aimed at illicitly extracting Claude's capabilities.
Security Risks of Distilled Models
Anthropic cautioned that distilled models present unique security risks because they may lack the inherent safeguards of the original, larger models.
Protocol Stripping: These campaigns could allow actors to deliberately remove security protocols.
Bias Control: They could also undermine mechanisms designed to ensure AI models maintain ideological neutrality and truth-seeking behavior.
U.S. Government Response and Future Actions
The Trump administration plans a multi-faceted response to these alleged activities:
Information Sharing: Sharing intelligence with U.S. AI companies regarding unauthorized, industrial-scale distillation attempts.
Coordination: Improving coordination between private technology companies.
Accountability: Exploring measures to hold foreign actors responsible for such campaigns.
Best Practices: Working with the private sector to develop defenses against these activities.
These actions underscore the escalating technological competition between the U.S. and China, with AI technology becoming a major flashpoint in global trade tensions.
