Anthropic Alleges Chinese AI Labs Used 24,000 Fraudulent Accounts to Extract Claude Capabilities
China
Dado Ruvic/Reuters
What Happened
Why It Matters
What Happened
Why It Matters
Where Sources Agree
- arrows_inputAnthropic Alleges Claude Extraction: Most outlets confirm Anthropic accused three Chinese AI labs of illicitly extracting Claude's capabilities through 24,000 fraudulent accounts that generated 16 million exchanges, according to Anthropic.
- arrows_inputAI Distillation Security Risk: Coverage highlights that illicitly distilled models lack necessary safeguards, framing it as a significant national security risk, according to Anthropic.
- arrows_inputChinese Labs Remain Silent: Reporting predominantly indicates that Chinese AI labs DeepSeek, MiniMax, and Moonshot AI did not respond to requests for comment regarding Anthropic's allegations, according to multiple reports.
Where Sources Disagree
- arrows_outputIP Claims vs. Training History: Anthropic accuses Chinese AI labs of 'industrial-scale campaigns' to illicitly extract Claude's capabilities, yet critics accuse Anthropic of hypocrisy, citing its history of training Claude on copyrighted books and settling a lawsuit.
- arrows_outputAI Export Control Effectiveness: Anthropic argues that AI model distillation reinforces the rationale for US export controls, limiting both direct training and illicit capability extraction. However, some critics contend that distillation undermines these controls by enabling foreign labs to copy capabilities without direct chip access.
- arrows_outputDistillation Characterization: Anthropic characterizes Chinese labs' use of its Claude AI as 'illicit' and 'industrial-scale theft,' violating terms of service. However, some reports emphasize that distillation is a widely used and legitimate training method, noting the legal status of such 'illicit' use remains unclear.
Timeline
February 24, 2026
Detection, mitigation and policy push: Anthropic said it detected the campaigns using IP and metadata correlations, shared findings with U.S. government entities and industry, built classifiers and behavioral fingerprinting, strengthened account verification, and urged export controls and coordinated action—framing the attacks as a national security risk. The company also noted it intervened while some campaigns were still active and shared technical indicators.
February 24, 2026
Campaigns' scope and targets: Anthropic reported a breakdown showing MiniMax generated about 13 million exchanges, Moonshot about 3.4 million, and DeepSeek roughly 150,000, with prompts focused on agentic reasoning, tool use, coding, computer vision and creating censorship‑safe alternatives; the activity concentrated on Claude’s most advanced capabilities and unfolded over weeks to months.
February 24, 2026
Anthropic discloses extraction campaigns: On February 24, Anthropic announced it had uncovered "industrial-scale" distillation campaigns by three China-based labs—DeepSeek, Moonshot and MiniMax—that used roughly 24,000 fraudulent accounts to generate more than 16 million exchanges with its Claude chatbot in violation of terms and regional access restrictions. The company characterized the activity as illicitly extracting Claude's capabilities.
Summary by Ground AI
Sources
See All 91A varied selection of sources chosen by Ground to reflect the diversity of this story’s coverage.
Most Written About
The stories getting the most coverage from the last 24 hours
Timeline
February 24, 2026
Detection, mitigation and policy push: Anthropic said it detected the campaigns using IP and metadata correlations, shared findings with U.S. government entities and industry, built classifiers and behavioral fingerprinting, strengthened account verification, and urged export controls and coordinated action—framing the attacks as a national security risk. The company also noted it intervened while some campaigns were still active and shared technical indicators.
February 24, 2026
Campaigns' scope and targets: Anthropic reported a breakdown showing MiniMax generated about 13 million exchanges, Moonshot about 3.4 million, and DeepSeek roughly 150,000, with prompts focused on agentic reasoning, tool use, coding, computer vision and creating censorship‑safe alternatives; the activity concentrated on Claude’s most advanced capabilities and unfolded over weeks to months.
February 24, 2026
Anthropic discloses extraction campaigns: On February 24, Anthropic announced it had uncovered "industrial-scale" distillation campaigns by three China-based labs—DeepSeek, Moonshot and MiniMax—that used roughly 24,000 fraudulent accounts to generate more than 16 million exchanges with its Claude chatbot in violation of terms and regional access restrictions. The company characterized the activity as illicitly extracting Claude's capabilities.
