Allegations of Model Distillation Against Chinese AI Labs
Anthropic has released details regarding what it characterizes as large-scale distillation campaigns conducted by several prominent Chinese artificial intelligence laboratories. According to reports from TechCrunch, the company alleges that Alibaba, Moonshot AI, and DeepSeek utilized Anthropic’s Claude models to train their own proprietary systems through high-volume data exchanges.
Scale of Data Exchanges
The data released by Anthropic suggests a significant volume of interaction between the Chinese models and Claude’s infrastructure. The company reports that Alibaba Group Holding engaged in approximately 151 million exchanges over a 14-day period. Moonshot AI reportedly conducted 23 million exchanges, while DeepSeek engaged in 12 million exchanges during the same timeframe.
Furthermore, Anthropic alleges that Kimi and DeepSeek engaged in a practice of secretly serving Claude Opus to their own users to harvest “chain-of-thought” reasoning data. This technique, often referred to in the industry as distillation, involves using the outputs of a larger, more advanced model to train a smaller, more efficient one.
Industry Debate and Counterarguments
The allegations have sparked significant debate within the technical community. On social media platforms such as Reddit, some users have questioned the characterization of these activities as “attacks.” Critics argue that distillation is a standard industry technique used by many AI laboratories to optimize smaller models and that the labs in question were simply paying for tokens to receive responses as the models were trained to provide them.
Some commentators have suggested that the focus on Chinese firms may be influenced by geopolitical tensions. These observers note that while distillation may violate terms of service, similar practices regarding the use of copyrighted materials have been a point of contention for U.S.-based labs as well. Additionally, some analysts argue that the rapid advancement of Chinese AI is not solely due to distillation, pointing to the high volume of technical papers published by these labs as evidence of independent research and development.
Corporate Context and Uncertainty
The allegations arrive amid broader discussions regarding Anthropic’s corporate trajectory. While the company has previously stated it has implemented safeguards to block AI-assisted development of bioweapons, the company continues to face scrutiny regarding its financial requirements. Some market observers have questioned the necessity of Anthropic’s massive capital requirements and the timing of potential public offerings given the current macroeconomic uncertainty.
As of now, the specific technical methodologies used by Alibaba, Moonshot AI, and DeepSeek to achieve their current performance levels remain a subject of intense scrutiny, with the industry divided on whether these developments are the result of proprietary breakthroughs or the strategic use of distilled data.