OpenAI and Anthropic have warned U.S. officials that tens of thousands of fake accounts—allegedly tied to China—are being used to harvest proprietary capabilities from their systems through “distillation,” a low-cost technique for replicating model performance. The companies’ letters, echoing a White House memo this spring, describe the activity as industrial in scale and aimed at repackaging U.S. chatbot know-how as rival products. The administration has said it possesses evidence of systematic attempts to extract sensitive information and capabilities from American AI firms and has outlined steps to brief companies, harden defenses, develop best practices, and consider measures to hold foreign actors accountable. Executives say the activity has intensified, underscoring both the commercial stakes in the U.S.-China AI race and the challenge of policing model theft through public APIs and disposable accounts.
Related articles:
— AI Risk Management Framework (NIST)
— Artificial Intelligence: Cybersecurity Challenges (ENISA)
— OECD Principles on Artificial Intelligence
