Chinese AI Firms Accused of Stealing Anthropic’s Claude Chatbot Capabilities

February 23, 2026 – US artificial intelligence firm Anthropic has accused three Chinese companies of engaging in a large-scale effort to illicitly acquire its proprietary technology, raising concerns about intellectual property theft and potential national security implications. The allegations, mirroring similar claims leveled against Chinese firms by OpenAI earlier this month, center on the practice of “distillation,” a technique used to rapidly improve the performance of less capable AI models by leveraging the outputs of more advanced ones.

Distillation: A Common Practice, But With Risks

Anthropic alleges that DeepSeek, Moonshot AI, and MiniMax systematically extracted capabilities from its Claude chatbot through approximately 16 million interactions generated by roughly 24,000 fraudulent accounts. While distillation itself is a recognized technique within the AI development community – often employed to create smaller, more cost-effective versions of existing models – Anthropic argues that the scale and coordinated nature of these campaigns represent a deliberate attempt to circumvent export controls and gain an unfair competitive advantage.

The company’s concerns extend beyond simple intellectual property theft. Anthropic contends that models built through this method are less likely to retain the crucial safety guardrails embedded in leading US AI systems. These guardrails are designed to prevent misuse of the technology, including applications in areas like weapons development and cyber warfare. The potential for these safeguards to be stripped away raises significant national security concerns, according to the company.

Circumventing Restrictions and Scaling the Attack

To bypass Anthropic’s direct restrictions on commercial access from China, the accused firms allegedly utilized proxy services to manage the sprawling networks of fraudulent accounts. This allowed them to distribute traffic across Anthropic’s API and third-party cloud platforms, masking the origin of the requests and enabling a sustained, large-scale data extraction effort. Anthropic identified the activity through analysis of IP address correlations, request metadata, and infrastructure indicators that deviated significantly from typical user traffic.

MiniMax reportedly spearheaded the largest operation, generating over 13 million exchanges with Claude. Anthropic noted that all three campaigns focused heavily on Claude’s most advanced capabilities – complex reasoning, coding, and tool use – areas where the chatbot is considered a leader in the field. This targeted approach suggests a deliberate effort to acquire specific, high-value skills rather than simply gathering general data.

A Broader Trend and Calls for Coordinated Action

The accusations against DeepSeek, Moonshot AI, and MiniMax follow similar charges leveled against DeepSeek by OpenAI last month. This suggests a broader trend of Chinese firms attempting to leverage US AI technology through indirect means. The emergence of DeepSeek’s low-cost generative AI model last year, which achieved performance levels comparable to ChatGPT and other leading US chatbots, underscored the potential for rapid advancements in the sector and raised questions about the continued dominance of American companies.

Anthropic emphasized that addressing this challenge requires a coordinated response involving industry players, policymakers, and the global AI community. The company argues that no single entity can effectively tackle the issue alone, given its complexity and scale. The timing of these accusations comes as Washington continues to tighten export controls aimed at preserving America’s technological edge in the field of artificial intelligence.

Implications for the AI Landscape

The allegations highlight the growing competition in the AI sector and the lengths to which companies are willing to go to gain an advantage. The practice of distillation, while not inherently malicious, raises ethical and legal questions when employed on such a large scale and with the apparent intent of circumventing intellectual property rights and export controls.

The incident also underscores the importance of robust security measures and monitoring systems to detect and prevent unauthorized access to sensitive AI models. As AI technology becomes increasingly powerful and pervasive, protecting intellectual property and ensuring responsible development will be critical to maintaining trust and fostering innovation. The lack of safety guardrails in distilled models, as highlighted by Anthropic, presents a potential risk that could have far-reaching consequences.

The situation is likely to intensify scrutiny of technology transfer and intellectual property protection within the AI industry, potentially leading to further restrictions and increased enforcement efforts. It also raises questions about the effectiveness of current export control measures and the need for international cooperation to address the challenges posed by rapidly evolving AI technologies.