In a significant upheaval within the artificial intelligence industry, Anthropic, a leading AI company based in the United States, has leveled serious allegations against three Chinese firms: DeepSeek, Moonshot AI, and MiniMax. The accusation involves these companies having engaged in illegal extraction of AI capabilities from Anthropic’s Claude chatbot through a method called “distillation.” This technique involves using the outputs from advanced AI systems to improve the functionality of less capable ones, and when misused, it can lead to significant ethical and ownership concerns.
These claims by Anthropic are not isolated; they echo similar concerns raised by OpenAI the previous month, highlighting a worrisome escalation in thefts of intellectual property in the AI sector. Anthropic asserts that these illicit activities are increasingly sophisticated, threatening proprietary technologies and posing potential risks to national security.
To illustrate the scale of the alleged theft, Anthropic points out that the accused firms engaged in approximately 16 million interactions with the Claude model, using around 24,000 fraudulent accounts to extract capabilities they did not develop. The utilization of distillation, while a common practice in AI development for producing smaller, cost-effective model versions, raises significant concerns when it’s used to appropriate foreign technological advancements without consent or compensation.
For instance, DeepSeek, one of the implicated companies, introduced a low-cost generative AI model last year that reportedly performs as well as top American competitors like ChatGPT. This development challenges the technological supremacy of the US in AI—a concern that underscores the potential impact of stolen AI technology.
In response to these allegations, Anthropic has appealed for a unified strategy from both the industry and governmental bodies to prevent the misuse of distillation and to enforce export controls rigorously. The company warns that models developed through allegedly unlawful means might not include crucial safety protocols, posing risks of misuse in areas such as cyberattacks or even the manufacturing of bioweapons.
Particularly alarming is the claim that MiniMax, the largest offender, orchestrated over 13 million interactions focusing on sophisticated domains such as coding and agentic reasoning, where Claude has shown superiority. The accused firms reportedly circumvented Anthropic’s security measures by channeling their efforts through proxy services and a web of fake accounts.
This series of events calls for immediate attention and action. As AI continues to grow in influence and application, it is apparent that more stringent regulatory frameworks and cooperative industry measures are needed to curb the ethical and legal challenges posed by advanced technologies. Ensuring the security and integrity of AI advancements requires a reevaluation of how distillation and similar techniques are monitored and managed, safeguarding against unauthorized exploitation.