Anthropic Accuses DeepSeek, Moonshot AI, and MiniMax of 'Industrial-Scale' Data Theft from Claude

Anthropic, the San Francisco-based AI safety company that develops the Claude chatbot, has publicly accused three of China's most prominent AI startups — DeepSeek, Moonshot AI, and MiniMax — of conducting what it describes as 'industrial-scale intellectual property theft.' The allegations, disclosed in February 2026, center on a technique called distillation: systematically querying a more advanced AI model to extract its capabilities and use them to boost a less capable competitor.

According to Anthropic's disclosure, the three companies collectively conducted approximately 16 million exchanges with Claude through about 24,000 fraudulent accounts, violating both Anthropic's terms of service and regional access restrictions that were designed to prevent exactly this kind of capability extraction. The campaigns, Anthropic warned, are growing in both intensity and sophistication.

How AI Distillation Works — and Why It Threatens the Industry

Model distillation is not new in machine learning. The technique, originally described by Geoffrey Hinton in 2015, involves training a smaller 'student' model to mimic the output patterns of a larger 'teacher' model. In legitimate applications, companies use distillation to create compact versions of their own models for deployment on edge devices or in cost-sensitive environments.

What Anthropic alleges is distillation used as a weapon — querying a competitor's model millions of times to create training data that captures its most advanced reasoning capabilities, coding ability, and tool-use skills. The student model, having learned to replicate the teacher's outputs, can then demonstrate capabilities that would have taken years of independent research and billions of dollars in compute to develop from scratch.

graph TD
    A["Attacker creates 24,000 fake accounts"] --> B["Automated scripts query Claude 16M times"]
    B --> C["Responses captured as training data"]
    C --> D["Student model trained on Claude outputs"]
    D --> E["Student model gains Claude's capabilities"]
    E --> F["Deployed commercially without licensing"]
    style A fill:#ff6b6b,color:#fff
    style F fill:#ff6b6b,color:#fff

The scale of the alleged operation — 16 million exchanges — is staggering. To put this in perspective, a single human user typing continuously for 8 hours a day would take approximately 22 years to generate that many interactions. The automated nature of the campaign suggests purpose-built infrastructure designed specifically for high-volume capability extraction.

The Three Accused Companies

DeepSeek, the Hangzhou-based AI lab that stunned the industry in January 2025 with its open-weight DeepSeek-V3 model, is the highest-profile company named. DeepSeek gained attention for matching or exceeding Western AI models at a fraction of the reported training cost. Anthropic's allegations suggest that at least some of DeepSeek's rapid performance improvements may have been achieved through illicit distillation rather than independent innovation.

Moonshot AI, whose Kimi chatbot has become one of China's most popular AI assistants, and MiniMax, which develops the M2.5 model that TensorVue recently covered as challenging Claude Opus at a fraction of the cost, were also named. Both companies have been notable for the speed at which their models have improved, particularly in domains like coding and agentic reasoning — exactly the capabilities that Anthropic says were targeted in the distillation campaign.

National Security Dimensions

Anthropic framed its allegations partly in national security terms, arguing that models built through illicit distillation may lack the safety guardrails that American AI companies invest heavily in developing. When a student model copies a teacher's capabilities, it does not necessarily copy its safety training — the carefully engineered constraints that prevent AI models from generating harmful content, assisting with weapons development, or enabling surveillance.

This safety gap creates a paradox: the distilled models may be more capable but also more dangerous, combining advanced reasoning with fewer ethical constraints. Anthropic argued that this represents a direct circumvention of US export controls designed to maintain American leadership in AI technology — not by exporting chips or models, but by extracting the functional equivalent of years of American AI research through API abuse.

Industry-Wide Problem: OpenAI Reports Similar Attacks

Anthropic is not alone in making these accusations. OpenAI had separately informed US lawmakers earlier in February 2026 about ongoing efforts by Chinese companies to leverage capabilities developed by American AI labs. The parallel disclosures suggest a coordinated industry campaign to draw regulatory attention to what both companies view as a systemic threat.

The timing is significant. Both Anthropic and OpenAI face competitive pressure from Chinese models that are increasingly matching their performance at lower prices. By framing this competition as intellectual property theft rather than legitimate market competition, the American companies may be seeking regulatory protections that would limit API access from certain countries or require more rigorous identity verification for API users.

What Happens Next

Anthropic has called for a coordinated industry and government response, including stricter API access controls, mandatory identity verification for high-volume users, and potential legislative action to classify AI distillation attacks as a form of trade secret theft. The companies are also likely to invest in technical countermeasures — watermarking model outputs, rate-limiting suspicious accounts, and using honeypot queries to detect distillation attempts.

The accused companies have not issued detailed public responses to the allegations. The geopolitical dimensions of the dispute — touching on trade tensions, technology competition, and national security — ensure that this controversy will extend well beyond the AI industry into the broader arena of US-China technological rivalry.