In a stunning revelation that has sent shockwaves through the global AI industry, Anthropic, the developer of the advanced Claude AI model, has publicly accused three prominent Chinese AI companies of orchestrating large-scale intellectual property theft. According to Anthropic’s official blog post released on February 23, 2026, companies DeepSeek, Moonshot AI, and MiniMax allegedly created over 24,000 fraudulent accounts to generate more than 16 million exchanges with Claude. This industrial-scale operation, described as “distillation attacks,” aimed to illicitly extract and replicate Claude’s advanced capabilities to boost their own models—bypassing U.S. export controls and terms of service restrictions.
This incident marks a dramatic escalation in what many are calling the “AI war”—a fierce geopolitical and technological rivalry between U.S. and Chinese AI developers. As nations race to dominate artificial intelligence, such tactics highlight the high stakes involved, from national security risks to the erosion of innovation incentives.
What Exactly Happened? Breaking Down Anthropic’s Accusations
Anthropic’s detailed report, titled “Detecting and Preventing Distillation Attacks,” outlines how these Chinese labs allegedly operated. Distillation is a technique where outputs from a more powerful “teacher” model (in this case, Claude) are used to train a smaller, less capable “student” model. It’s an efficient way to transfer knowledge without building everything from scratch—but when done without permission, it’s essentially theft of proprietary technology.
The playbook was remarkably consistent across the three companies:
- Fraudulent Account Creation: Using proxy services and automated tools, the labs set up thousands of fake accounts to circumvent Anthropic’s regional bans. Claude is not commercially available in China due to legal, regulatory, and national security concerns.
- Massive Query Volume: Over 16 million interactions were logged, far exceeding normal user patterns. This allowed the attackers to probe Claude’s strengths in key areas like agentic reasoning (autonomous decision-making), tool use, coding, and data analysis.
- Targeted Extraction: Specific campaigns focused on Claude’s differentiated features. For instance:
- DeepSeek reportedly ran over 150,000 exchanges aimed at improving foundational logic and alignment, including censorship-safe responses to sensitive queries.
- Moonshot AI generated more than 3.4 million exchanges targeting agentic reasoning, tool use, coding agents, and even computer vision.
- MiniMax’s efforts were similarly scaled to replicate advanced functionalities.
Anthropic emphasized that these actions violated its terms of service and regional access restrictions. By evading detection through proxies and account rotation, the campaigns achieved “industrial-scale” efficiency at a fraction of the cost of independent development.
This isn’t isolated. Similar allegations have surfaced before—OpenAI accused Chinese firms of comparable tactics just last month—pointing to a broader pattern of reverse-engineering frontier U.S. models.
The Bigger Picture: Why This Is a National Security and Geopolitical Issue
The U.S. government has long imposed strict export controls on advanced AI chips and technologies to prevent adversaries from closing the capability gap. Tools like NVIDIA’s high-end GPUs are restricted from sale to China, aiming to slow the development of rival frontier models.
However, distillation attacks offer a workaround: access powerful models indirectly through APIs, harvest high-quality outputs, and use them to bootstrap domestic systems. Anthropic warns that distilled models often lack proper safeguards, potentially amplifying risks like misalignment, harmful outputs, or misuse in cyber operations.
This ties into broader concerns. In late 2025, Anthropic disrupted what it called the first AI-orchestrated cyber espionage campaign, attributed with high confidence to a Chinese state-sponsored group using Claude in attacks on global targets. While separate, these incidents fuel fears that unchecked access could enable malicious applications.
Critics have noted irony—some point out that AI companies like Anthropic and OpenAI have faced lawsuits (e.g., from Reddit) over unauthorized data scraping for training. Yet the scale and deliberate circumvention here set this apart as targeted IP theft.
Who Are the Accused Chinese AI Companies?
- DeepSeek: A rising player known for cost-effective, high-performance models. It has gained attention for open-source releases that rival Western counterparts.
- Moonshot AI: Backed by significant funding, Moonshot focuses on advanced chatbots and agents. Its Kimi series has been praised for strong reasoning and coding abilities.
- MiniMax: Specializes in multimodal and competitive frontier models, claiming parity with top industry offerings.
All three are Chinese-based and operate in a ecosystem where government support for AI is intense. China aims for AI leadership by 2030, investing heavily in domestic alternatives amid U.S. restrictions.
Implications for the Global AI Landscape
This revelation underscores several key trends:
- Intensifying AI Competition: The gap between U.S. and Chinese capabilities is narrowing faster than expected, partly through such methods.
- Need for Stronger Defenses: Anthropic is enhancing detection, including better monitoring of anomalous patterns, rate limiting, and proxy blocking.
- Policy Ramifications: Calls for tighter API controls, international agreements on AI IP, and perhaps expanded export rules are growing louder.
- Innovation Risks: If frontier models can be cheaply replicated, it disincentivizes massive R&D investments.
For businesses and developers relying on Claude or similar tools, this highlights the importance of ethical AI use and robust terms enforcement.
What Happens Next in the AI War?
Anthropic’s post ends with a stark warning: “The window to act is narrow.” As campaigns grow in sophistication, the industry must collaborate on defenses—perhaps through shared threat intelligence or advanced watermarking of outputs.
Whether this leads to diplomatic tensions, legal actions, or accelerated U.S. countermeasures remains to be seen. One thing is clear: the battle for AI supremacy is no longer just about compute power—it’s about protecting the crown jewels of intelligence itself.
At VFutureMedia, we track the latest in AI, tech geopolitics, and digital innovation. Stay tuned for more analysis on how events like this shape the future of artificial intelligence.
Leave a Comment