Anthropic Accuses Chinese AI Firms of Mass Data Harvesting as US Confirms DeepSeek Used Restricted Nvidia Chips

Anthropic has accused Chinese firms DeepSeek, Moonshot, and MiniMax of harvesting 16 million conversations from its Claude models to train their own systems. Simultaneously, US officials confirmed DeepSeek utilised restricted Nvidia Blackwell chips, smuggled into China, intensifying concerns over the effectiveness of current American technology export controls.

Anthropic AI (Photo Credits: Official Website)

San Francisco, February 24: American artificial intelligence safety start-up Anthropic has accused three prominent Chinese companies of conducting a massive "distillation" campaign to harvest proprietary data from its Claude models. The company alleges that DeepSeek, Moonshot, and MiniMax utilised over 24,000 fraudulent accounts to generate more than 16 million exchanges, aimed at extracting advanced reasoning and coding capabilities for their own AI developments.

The allegations surface as United States government officials confirmed that DeepSeek successfully trained its latest AI models using Nvidia’s highly restricted Blackwell chips. Despite a federal ban on shipping these advanced semiconductors to China, the chips were reportedly operated at a facility in Inner Mongolia. This development has triggered a fresh debate in Washington regarding the circumvention of export controls and the narrowing window to protect American intellectual property. Payal Gaming aka Payal Dhare Meets Sundar Pichai Months After 'Viral MMS Video' Controversy, Shares Photos.

Industrial-Scale Data Extraction

Anthropic reported that the volume and structure of the queries from the Chinese firms reflected "deliberate capability extraction" rather than legitimate user interaction. According to a blog post released on Monday, MiniMax accounted for 13 million exchanges, while Moonshot recorded 3.4 million. DeepSeek was involved in approximately 150,000 exchanges specifically designed to force Claude to reveal its internal "chain-of-thought" logic.

The San Francisco-based lab noted that these campaigns targeted Claude’s most sophisticated features, including agentic reasoning and tool orchestration. Furthermore, Anthropic observed tasks where Claude was prompted to generate censorship-safe responses to politically sensitive topics involving dissidents or party leaders, suggesting the data was intended to help Chinese models navigate domestic censorship requirements.

Circumvention of US Hardware Restrictions

While Anthropic highlighted the theft of "software" logic, US officials provided confirmation to Reuters that DeepSeek also bypassed hardware restrictions. The Hangzhou-based company reportedly used smuggled Nvidia Blackwell chips—the most advanced AI processors currently available—to achieve performance levels that rival top-tier American models.

The US Commerce Department currently prohibits the shipment of Blackwell chips to China to prevent them from being used for military advancements. However, the presence of these chips in Inner Mongolia suggests a robust black market or "grey market" for high-end silicon. Analysts suggest that the combination of smuggled hardware and distilled data from US models like Claude and ChatGPT has allowed Chinese firms to close the technological gap at a fraction of the traditional cost.

A Growing Policy Divide in Washington

The revelation of DeepSeek's hardware acquisition has intensified the divide among US policymakers. White House AI Czar David Sacks and Nvidia CEO Jensen Huang have previously argued that shipping some advanced chips to China might actually discourage domestic Chinese innovation by firms like Huawei. Conversely, national security hawks argue that any access to high-end chips threatens US dominance and aids the Chinese military. Grand Theft Auto VI Price in India: Rockstar Confirms November 19 Launch; Check Prices, Supported Consoles, Gameplay and More.

Anthropic warned that the "window to act is narrow" for the global AI community to address these sophisticated distillation attacks. The company called for rapid, coordinated action among industry players and policymakers to prevent the illicit acquisition of powerful AI capabilities. The Chinese embassy in Washington has responded by opposing the "politicisation" of technological issues, asserting that US export controls overstretch the concept of national security.

(The above story first appeared on LatestLY on Feb 24, 2026 12:20 PM IST. For more news and updates on politics, world, sports, entertainment and lifestyle, log on to our website latestly.com).

Share Now

Share Now