Anthropic Accuses Chinese AI Firms of Illegally Harvesting Claude Data

San Francisco, February 23, 2026 — Anthropic PBC has issued a formal condemnation of three prominent Chinese artificial intelligence firms, accusing them of systematically harvesting data from its Claude conversational AI model to accelerate the development of their own proprietary systems. In a detailed blog post released today, Anthropic named DeepSeek Ltd., Moonshot AI, and MiniMax as the primary actors in what it describes as a large-scale, coordinated effort to circumvent ethical and legal boundaries in AI training.

According to Anthropic’s internal analysis, these companies deployed automated scripts and proxy services to interact with Claude’s public-facing interfaces, collecting millions of user prompts, model responses, and contextual interactions. The data, Anthropic alleges, was then used to fine-tune their own large language models—bypassing the need for costly, time-intensive data collection and labeling processes. This practice, known in AI research as "model distillation," is not inherently illegal, but Anthropic contends that the scale, intent, and method of extraction violate its Terms of Service and potentially infringe on intellectual property rights.

"We built Claude with the explicit understanding that user interactions are protected and used solely to improve our own system," said a spokesperson for Anthropic in a statement. "When third parties systematically scrape and repurpose this data to build competing products, it undermines trust, distorts market competition, and puts user privacy at risk."

While the practice of reverse-engineering AI models through query-response analysis has been documented in academic circles, Anthropic’s allegations point to an unprecedented industrial-scale operation. Internal logs cited by the company show repeated, high-frequency queries originating from IP addresses linked to the accused firms, often with patterns indicative of automated bots rather than human users. The data harvested reportedly included sensitive user inputs—ranging from personal advice requests to confidential business inquiries—that were never intended for public reuse.

DeepSeek, Moonshot, and MiniMax have not issued formal public responses to the allegations as of press time. However, industry insiders suggest the companies may be leveraging open-source techniques to mitigate legal exposure. "These firms operate in a regulatory gray zone," said Dr. Lena Zhao, an AI ethics researcher at Tsinghua University. "China’s current AI governance framework focuses more on national security and content control than on proprietary data protection. That creates a structural incentive for firms to exploit loopholes in foreign platforms’ terms."

Legal experts warn the case could set a precedent for cross-border AI regulation. "If proven, this isn’t just a breach of contract—it’s a potential violation of the Digital Millennium Copyright Act and possibly the Computer Fraud and Abuse Act," said Professor Mark Ellison of Stanford Law School. "The U.S. government may be compelled to act if U.S.-based AI infrastructure is being weaponized as a free training dataset by foreign competitors."

Anthropic has reportedly begun implementing technical countermeasures, including rate-limiting, CAPTCHA challenges, and behavioral anomaly detection, to block suspicious access patterns. The company is also exploring legal avenues, including potential litigation under international trade and intellectual property law.

The incident underscores a growing tension in the global AI ecosystem: as Chinese firms rapidly close the performance gap with Western counterparts, accusations of data appropriation are becoming more frequent. While innovation is often built on existing knowledge, Anthropic’s case raises critical questions about where legitimate learning ends and unethical exploitation begins.

As the world’s largest AI models become increasingly valuable assets, the battle over data sovereignty may define the next decade of technological competition. For now, the burden falls on developers and regulators to define the rules of engagement before the AI arms race spirals beyond ethical bounds.