<figure class="wp-block-image size-large"><img width="325px" height="183px" srcSet="https://d2ly46a6b1u3g6.cloudfront.net/wp-content/uploads/2026/02/09114140/News-Debate-Exposes-Deep-Divide-on-AGIs-Long-Term-Consequences-1-375x212.webp 375w, https://d2ly46a6b1u3g6.cloudfront.net/wp-content/uploads/2026/02/09114140/News-Debate-Exposes-Deep-Divide-on-AGIs-Long-Term-Consequences-1-662x374.webp 662w, https://d2ly46a6b1u3g6.cloudfront.net/wp-content/uploads/2026/02/09114140/News-Debate-Exposes-Deep-Divide-on-AGIs-Long-Term-Consequences-1.webp 686w" sizes="(max-width: 425px) 150px, 450px" src="https://ll-blog-dev.s3.ap-southeast-2.amazonaws.com/wp-content/uploads/2026/02/09114140/News-Debate-Exposes-Deep-Divide-on-AGIs-Long-Term-Consequences-1-662x374.webp" alt="" class="wp-image-9210"/></figure>


Anthropic says three Chinese AI companies stole its tech. The accusation: large-scale &quot;distillation attacks&quot; against Claude.
The companies? DeepSeek, Moonshot, and MiniMax.
According to <a href="https://www.anthropic.com/news/detecting-and-preventing-distillation-attacks" target="_blank" rel="noopener">Anthropic&#x27;s blog post</a>, the firms created roughly 24,000 fake accounts. They generated over 16 million exchanges with Claude. Purpose: train their own systems without permission.
The attacks weren&#x27;t random. They targeted Claude&#x27;s &quot;most differentiated capabilities.&quot; That&#x27;s agentic reasoning, tool use, coding, data analysis, rubric-based grading, and computer vision.
Distillation itself isn&#x27;t new. It&#x27;s standard optimization. You teach a smaller AI model using outputs from a more advanced one. Companies do this internally all the time.
But Anthropic says this is different. Using distillation against another company&#x27;s model lets competitors skip years of expensive R&amp;D. Years they didn&#x27;t pay for.
How did Anthropic identify the three firms? IP address correlations. Request metadata. Infrastructure indicators. Plus confirmation from industry partners who saw similar activity.
All three accused companies are China-based AI startups. They&#x27;ve got multibillion-dollar valuations. DeepSeek&#x27;s already got significant international recognition.
Anthropic doesn&#x27;t call this isolated misuse. They&#x27;re calling it &quot;organized, industrial-scale scraping aimed at replicating frontier AI capabilities.&quot;
The stakes go beyond corporate IP disputes. Anthropic warns that foreign distillation of American AI models creates geopolitical risks. Security risks.
Extracted capabilities could be channeled into military operations. Intelligence gathering. Cyber activities. Disinformation campaigns. Mass surveillance by authoritarian governments.
The scale tells the story. Tens of thousands of fake accounts. Millions of interactions. This wasn&#x27;t casual API usage. It wasn&#x27;t standard competitive research. It was systematic.
Anthropic&#x27;s response? They&#x27;re strengthening detection systems. Tightening access controls. Sharing threat intelligence with other organizations.
The company&#x27;s calling for coordinated action. They want AI labs involved. Cloud providers. Policymakers. Goal: combat large-scale distillation attacks and protect frontier AI capabilities from unauthorized extraction.

Kalshi Voids Trades on Khamenei's Death, Citing Policy

Backpack's Token-to-Equity Model Seeks to Sidestep Securities Regulations

Congress Moves to Provide Legal Protections for Blockchain Developers

SBF's Prison Endorsement Backfires: Lummis &amp; Warren Unite to Slam 'Toxic Support' for CLARITY Act

<figure class="wp-block-image size-large"><img width="325px" height="183px" srcset="https://d2ly46a6b1u3g6.cloudfront.net/wp-content/uploads/2026/02/09114140/News-Debate-Exposes-Deep-Divide-on-AGIs-Long-Term-Consequences-1-375x212.webp 375w, https://d2ly46a6b1u3g6.cloudfront.net/wp-content/uploads/2026/02/09114140/News-Debate-Exposes-Deep-Divide-on-AGIs-Long-Term-Consequences-1-662x374.webp 662w, https://d2ly46a6b1u3g6.cloudfront.net/wp-content/uploads/2026/02/09114140/News-Debate-Exposes-Deep-Divide-on-AGIs-Long-Term-Consequences-1.webp 686w" sizes="(max-width: 425px) 150px, 450px" src="https://ll-blog-dev.s3.ap-southeast-2.amazonaws.com/wp-content/uploads/2026/02/09114140/News-Debate-Exposes-Deep-Divide-on-AGIs-Long-Term-Consequences-1-662x374.webp" alt="" class="wp-image-9210" /></figure>


Anthropic says three Chinese AI companies stole its tech. The accusation: large-scale "distillation attacks" against Claude.
The companies? DeepSeek, Moonshot, and MiniMax.
According to <a href="https://www.anthropic.com/news/detecting-and-preventing-distillation-attacks" target="_blank" rel="noopener">Anthropic's blog post</a>, the firms created roughly 24,000 fake accounts. They generated over 16 million exchanges with Claude. Purpose: train their own systems without permission.
The attacks weren't random. They targeted Claude's "most differentiated capabilities." That's agentic reasoning, tool use, coding, data analysis, rubric-based grading, and computer vision.
Distillation itself isn't new. It's standard optimization. You teach a smaller AI model using outputs from a more advanced one. Companies do this internally all the time.
But Anthropic says this is different. Using distillation against another company's model lets competitors skip years of expensive R&amp;D. Years they didn't pay for.
How did Anthropic identify the three firms? IP address correlations. Request metadata. Infrastructure indicators. Plus confirmation from industry partners who saw similar activity.
All three accused companies are China-based AI startups. They've got multibillion-dollar valuations. DeepSeek's already got significant international recognition.
Anthropic doesn't call this isolated misuse. They're calling it "organized, industrial-scale scraping aimed at replicating frontier AI capabilities."
The stakes go beyond corporate IP disputes. Anthropic warns that foreign distillation of American AI models creates geopolitical risks. Security risks.
Extracted capabilities could be channeled into military operations. Intelligence gathering. Cyber activities. Disinformation campaigns. Mass surveillance by authoritarian governments.
The scale tells the story. Tens of thousands of fake accounts. Millions of interactions. This wasn't casual API usage. It wasn't standard competitive research. It was systematic.
Anthropic's response? They're strengthening detection systems. Tightening access controls. Sharing threat intelligence with other organizations.
The company's calling for coordinated action. They want AI labs involved. Cloud providers. Policymakers. Goal: combat large-scale distillation attacks and protect frontier AI capabilities from unauthorized extraction.

Anthropic Accuses China of Straight-Up AI Model Theft

Anthropic claims DeepSeek, Moonshot, and MiniMax used 24,000 fake accounts to steal Claude AI capabilities through distillation attacks. Geopolitical concerns.