In recent revelations, US-based AI firm Anthropic has uncovered a significant breach in security orchestrated by three Chinese companies: DeepSeek, Moonshot AI, and MiniMax. These firms leveraged an industrial-scale technique known as “distillation,” utilizing Claude, Anthropic’s chatbot, to rapidly enhance the capabilities of less sophisticated models, often at an exorbitant cost.
Anthropic’s statement emphasizes that these operations have escalated in sophistication over time. In a stark departure from established practices, DeepSeek and Moonshot AI employed this approach to skirt export restrictions on cutting-edge technology, aiming to maintain their competitive edge within the highly sensitive domain of artificial intelligence (AI).
The revelations surfaced as Distillation had previously garnered significant attention in 2022 when a low-cost generative AI model developed by DeepSeek achieved parity with top American chatbots like Anthropic’s Claude. At that time, it upended expectations regarding US dominance and underscored the vulnerability of certain technological advancements.
This latest breach has prompted Anthropic to call for coordinated action at both industry and governmental levels. The company asserts that models generated through illicit distillation are less likely to adhere to safety protocols designed to prevent their misuse — safeguards such as prohibiting the development of bioweapons or enabling cyberattacks.
At a similar scale, MiniMax operated, amassing over 13 million exchanges with Claude. Each campaign centered on areas where Claude is renowned for its proficiency: coding, agency reasoning, and tool use. The Chinese companies circumvented Anthropic’s restrictions by channeling traffic through extensive networks of fake accounts managed via proxy services.
Anthropic’s stance highlights the urgent need to address this cybersecurity threat comprehensively, recognizing that no single company can tackle it alone. This revelation underscores the necessity for robust international cooperation in safeguarding AI advancements and maintaining technological superiority against potential adversaries.