Background

Anthropic, the creator of the Claude series of large‑language models, announced that it had identified a coordinated effort by three Chinese AI firms—DeepSeek, MiniMax and Moonshot—to illicitly copy its technology. According to the company, the effort involved the creation of roughly 24,000 fraudulent accounts that were used to conduct more than 16 million interactions with Claude. These interactions were intended to extract the model’s reasoning capabilities and other advanced functions.

Allegations of Illicit Distillation

The three firms are accused of “distilling” Claude, a process that trains a smaller model using the outputs of a larger, more capable one. While Anthropic acknowledges that distillation can be a legitimate training method, it stresses that the technique can also be exploited for illicit purposes. By copying powerful capabilities at a fraction of the time and cost required for independent development, the firms allegedly bypassed the safeguards built into Claude.

DeepSeek, which has drawn attention for producing efficient yet powerful models, is reported to have conducted over 150,000 exchanges with Claude, specifically targeting its reasoning abilities. The company is also alleged to have used Claude to generate “censorship‑safe alternatives to politically sensitive questions about dissidents, party leaders, or authoritarianism.” MiniMax and Moonshot are said to have logged more than 13 million and 3.4 million exchanges, respectively.

Potential Risks and Implications

Anthropic warns that models distilled without the original safety layers could be repurposed for military, intelligence and surveillance applications. The firm argues that foreign labs that distill American models may feed these unprotected capabilities into cyber‑offensive operations, disinformation campaigns and mass‑surveillance systems, thereby empowering authoritarian governments.

The accusations echo similar concerns raised by OpenAI, which recently sent a letter to lawmakers alleging that DeepSeek was attempting to “free‑ride” on the capabilities developed by U.S. frontier labs.

Response and Calls for Action

In response to the findings, Anthropic is urging other AI industry participants, cloud service providers and policymakers to address the issue of illicit distillation. The company suggests that restricting access to advanced chips could limit the scale of unauthorized model training.

Anthropic’s announcement marks a significant escalation in the debate over how emerging AI technologies are protected, shared and regulated across borders. The firm’s call for coordinated industry and governmental action reflects growing anxiety about the potential for frontier AI models to be weaponized when safety mechanisms are stripped away.

Used: News Factory APP - news discovery and automation - ChatGPT for Business