In recent developments that have sparked significant debate in the AI community, Anthropic has accused operators affiliated with Alibaba and its AI lab of conducting one of the largest known distillation campaigns against its Claude models. According to CNBC, this alleged operation ran from April 22 to June 5, generating over 28.8 million interactions with Claude through approximately 25,000 fraudulent accounts.

This scale places the campaign in a different category compared to previous incidents. In February, Anthropic had previously named three Chinese AI labs—DeepSeek, Moonshot AI, and MiniMax—as having collectively generated more than 16 million Claude interactions through around 24,000 fraudulent accounts. The alleged Alibaba campaign dwarfs that combined total in just six weeks.

How 25,000 Fake Accounts Extracted Claude’s Core Intelligence

The process of distillation is a method used to transfer knowledge from one model to another. In this case, the campaign involved sending large volumes of carefully designed prompts to the target model and capturing its responses. These responses then became training data for the competing model, allowing it to learn and replicate the original model’s reasoning and responses without incurring the costs associated with developing such capabilities.

Detection of these distillation queries is challenging because they often appear identical to legitimate requests. A developer asking for help debugging a function and a campaign systematically extracting coding behavior can both send similar prompts. The only distinguishing factor is pattern recognition: massive volume, repetitive structures, and prompts targeting specific narrow capabilities arriving from hundreds of coordinated accounts in sequence.

The Safety Dimension Beyond Commercial Concerns

Beyond the commercial implications, there are significant safety concerns associated with unauthorized distillation of frontier models. When a lab distills a model without permission, it does not inherit the safety guardrails that were built into the original model. This means that dangerous capabilities can be transferred through the outputs, potentially leading to harmful outcomes.

While distillation itself is a legitimate and widely used technique for compressing large models into smaller, faster versions that run more efficiently, the line drawn by Anthropic is between using it on their own models (which is standard practice) and using it on a competitor’s model without permission. This distinction highlights the ethical and legal implications of such actions.

Anthropic Wants Congress to Make Model Theft Illegal

In response to these allegations, Anthropic’s Head of Policy, Sarah Heck, wrote a letter to senators expressing concerns about the illicit, systematic, and industrial-scale nature of these attacks. She emphasized that the activities were aimed at harvesting U.S. AI capabilities across frontier labs and repackaging them as their own without incurring the training and R&D costs involved.

House Republicans are also seeking sanctions on Chinese companies that copy American-made AI models. Sen. Bill Hagerty and Sen. Andy Kim are moving to add an amendment to defense legislation that would blacklist or sanction entities found conducting such campaigns, according to CNBC. The White House Office of Science and Technology Policy issued a memorandum in April warning of industrial-scale foreign distillation of U.S. AI models.

The Structural Problem Beyond Any Single Campaign

The structural problem extends beyond any single campaign. A distillation query is indistinguishable from a legitimate one, making it difficult to detect without advanced monitoring systems. The only way to fully close this gap is to restrict access to the model, which conflicts directly with the commercial logic of selling AI as a service.

If adversarial distillation becomes routine, AI labs may find themselves spending as much on access controls and identity verification as they do on training. This shift would treat every API call as a potential intelligence transfer rather than a revenue event. The implications for the future of AI development and security are significant, highlighting the need for robust measures to protect intellectual property and ensure ethical practices in model development.

Conclusion

The allegations against Alibaba and its affiliated operators highlight the growing concerns surrounding unauthorized distillation of AI models. As these incidents continue to unfold, it is crucial for stakeholders to remain vigilant and proactive in addressing the challenges posed by such activities. The future of AI development will depend on balancing innovation with ethical considerations and legal protections. Readers should keep an eye on developments in legislation and industry practices as they shape the landscape of AI security and intellectual property rights.

---

Original Source

This article is based on publicly available reporting. For the complete original story, visit the publisher’s article.