5Upper IntermediateExamTechnology

Anthropic Accuses Chinese AI Labs of Stealing Model Capabilities

AI company Anthropic has accused three Chinese laboratories of copying its technology through millions of secret exchanges. The case raises serious questions about intellectual property, safety, and competition in the artificial intelligence industry.

2 min readLevel 5February 24, 2026

By Flalingo Editorial Team

Vocabulary

Study these words before reading the article. Click the audio button to hear pronunciation.

distillation

/ˌdɪstɪˈleɪʃn/

noun

A process where a smaller AI model is trained using the outputs of a larger, more powerful model

“The technique of distillation allows companies to create cheaper versions of their best AI systems.”

illicit

/ɪˈlɪsɪt/

adjective

Not allowed by law or rules; illegal or forbidden

“The police discovered an illicit trade in stolen electronic goods.”

fraudulent

/ˈfrɔːdʒələnt/

adjective

Intended to deceive or trick someone, usually to gain money or an unfair advantage

“The bank closed several fraudulent accounts that had been used to steal money.”

safeguards

/ˈseɪfɡɑːrdz/

noun

Rules, systems, or actions designed to protect against danger or harm

“The government introduced new safeguards to protect people's personal data online.”

implications

/ˌɪmplɪˈkeɪʃnz/

noun

The possible effects or results of an action or decision

“The new law could have serious implications for small businesses across the country.”

coordinated

/kəʊˈɔːrdɪneɪtɪd/

adjective

Organized so that different parts work together in a planned and effective way

“The rescue effort was a carefully coordinated operation between several emergency teams.”

Listen to Article

0:00 / 0:00

Article

Click on any underlined word to see its definition. Highlighted words are from the vocabulary section.

On February 23, 2026, the American AI company Anthropic made a serious accusation. It claimed that three Chinese AI laboratories had conducted large-scale attacks against its technology. The companies named were DeepSeek, Moonshot AI, and MiniMax. According to Anthropic, these labs used a method called distillation to copy its AI model's abilities. The operation reportedly involved over 16 million exchanges through around 24,000 fake accounts.

Distillation is a technique where a weaker AI model learns from a stronger one. It is legitimate when companies use it on their own models. However, Anthropic argues it becomes illicit when competitors use it to steal capabilities. MiniMax was responsible for over 13 million exchanges targeting coding and tool use. Moonshot AI generated more than 3.4 million exchanges focused on reasoning and computer vision. DeepSeek conducted over 150,000 exchanges aimed at reasoning and avoiding political censorship.

Anthropic identified the campaigns through careful analysis of account activity. The company tracked patterns in data such as shared payment methods and coordinated timing. Attackers used networks of fraudulent accounts that Anthropic calls hydra clusters. If one account was banned, another would immediately replace it. Some accounts were even linked to the public profiles of senior laboratory staff.

This case has significant implications for national security and global AI competition. Anthropic warns that models built through illicit distillation often lack important safety safeguards. Without these protections, dangerous capabilities could spread to harmful actors around the world. The company believes this situation reinforces the need for stronger export controls on AI technology. Experts say that preventing such attacks will require cooperation across the entire AI industry.

Comprehension Quiz

Test your understanding of the article with these multiple-choice questions.

1
How many total exchanges did the three Chinese labs generate with Claude?
2
Which laboratory was responsible for the largest number of exchanges?
3
What is Anthropic's main concern about models built through illicit distillation?
4
How did Anthropic identify which labs were responsible for the attacks?
5
What was one specific goal of DeepSeek's distillation campaign?

Discussion

Answer these comprehension questions about the article.

1
According to the article, what is distillation and when does Anthropic consider it to be illicit?
2
The author suggests that these attacks have national security implications. What specific risks does Anthropic identify?
3
What evidence does the article provide to show how Anthropic identified the three laboratories responsible for the attacks?
4
According to the passage, which of the three Chinese laboratories was responsible for the largest volume of exchanges, and what did it target?
5
How does the article describe the 'hydra cluster' method used by the attackers to avoid detection?

Further Discussion

Share your opinions on these open-ended questions.

1
Do you think AI companies should be allowed to learn from each other's technology, or should there be strict rules against it? Why?
2
How do you think governments should balance the need for open scientific research with the need to protect intellectual property?
3
If you were leading an AI company, what measures would you take to protect your technology from being copied?
4
Some people argue that AI safety rules slow down progress. Do you agree or disagree with this view?
5
How might events like these distillation attacks change the relationship between countries that are competing in artificial intelligence?