AI Dispute Escalates as Anthropic Accuses Alibaba of Unauthorised Model Extraction

Anthropic, a US-based AI powerhouse, has accused Alibaba, a Chinese e-commerce and IT company, of "brazenly" and "illicitly" stealing the features of their Claude AI model. Claude and operators associated with Alibaba conducted about 29 million transactions, according to a letter delivered by the San Francisco-based business to two US senators.

It claimed to have carried out the biggest extraction campaign of its type, executing these trades using thousands of fake accounts. Congress should punish the corporations responsible for these kinds of attacks and increase security measures to stop the theft of US tech, according to Anthropic.

What Athropic’s Letter Stated?

In a letter dated 10 June, sent to US Senators Tim Scott and Elizabeth Warren, Anthropic accused Alibaba, a company listed on the New York Stock Exchange, of running the biggest operation to illegally steal Claude's skills. Anthropic claims that the campaign's backbone was a set of techniques called "distillation attacks" that used replies from a more robust AI model to train a less robust one. Claude's method of decision-making and its capacity to handle longer and more complicated jobs were allegedly targeted by operators associated with Alibaba, according to Anthropic.

Chinese corporations conduct these attacks on an "industrial scale" to steal American AI skills and pass them off as their own, the company claimed. Other purported assaults that Anthropic claimed endangered US forces were also mentioned in the letter. The company claimed that its geopolitical rivals receive a huge subsidy from the United States government when it launches distillation attacks, which drain hundreds of billions of dollars from American investment and R&D. According to the US Department of Defence, Alibaba and other large Chinese companies are involved with the Chinese military. This includes internet giant Baidu and automaker BYD.

Chinese Firm Accused of Targeting US AI Giants

All parties involved have refuted the claims, and Alibaba has taken legal action against the United States government this week in an effort to get its name removed from a Pentagon blacklist. American developers have levelled accusations against their Chinese counterparts, claiming that they train their models to compete with American AI technology at a tenth of the cost by employing distillation assaults.

Similar accusations have been levelled by OpenAI against Chinese groups in the past. Anthropic is preparing for a spectacular stock market debut, which might propel it to the ranks of the world's most valuable corporations. It will be joined by OpenAI, another prominent AI developer. However, Mythos and other more sophisticated Anthropic models have sparked cybersecurity worries due to their potential to exploit vulnerabilities in computer networks.

Bloomberg reports that OpenAI, Anthropic PBC, and Google, a subsidiary of Alphabet Inc., have started cooperating to prevent Chinese rivals from stealing data from state-of-the-art US AI models for competitive advantage.