In brief
U.S.-based AI startup Anthropic has publicly accused Chinese e‑commerce giant Alibaba of illicitly extracting information from its Claude model via a large-scale distillation attack. The company cited almost 29 million exchanges over fraudulent accounts as evidence. The letter, sent to Senators Tim Scott and Elizabeth Warren, urges Congress to penalise such theft and strengthen protections for American AI. Alibaba has denied the claims and this week sued the Pentagon to remove itself from a blacklist. The incident highlights growing concerns over Chinese firms reproducing U.S. tech through covert methods.
US artificial intelligence (AI) giant Anthropic has accused Chinese e‑commerce and technology firm Alibaba of brazenly and illicitly extracting its Claude model’s capabilities.
In a letter to two members of the U.S. Congress, the San Francisco‑based company claimed that operators linked to Alibaba conducted almost 29 million exchanges with Claude using thousands of fraudulent accounts. Anthropic described it as the largest extraction campaign of its kind.
The letter calls on Congress to penalise the entities responsible for such attacks and to intensify safeguards that prevent U.S. technology from being stolen. Anthropic also said the campaign risked national security, as it could provide enemy firms with advanced AI capabilities.
Alibaba has denied the allegations and this week filed a lawsuit against the U.S. government to remove its name from a Pentagon blacklist that links it to the Chinese military.
Anthropic’s claim centers on what the company calls “distillation attacks” – a method that extracts answers from a robust AI model to train a weaker one at scale. The company argues that this technique allows Chinese firms to harvest and repurpose U.S. AI expertise.
The U.S. Department of Defense has warned that Alibaba, as well as other Chinese tech firms like BYD and Baidu, may be tied to the Chinese military. The Anthropic letter cites such connections to bolster its argument that these attacks threaten U.S. military interests.
Previously, OpenAI and other American developers have alleged that Chinese competitors employ similar methods to train rival AI systems more cheaply. Anthropic’s own advanced models, including Mythos, have already raised security concerns about their potential to exploit computer system vulnerabilities.
With a potential blockbuster IPO on the horizon, Anthropic aims to become one of the world’s most valuable tech firms, but this latest controversy underscores the high stakes of AI innovation and intellectual property protection worldwide.
