Anthropic Accuses Chinese AI Labs of Secretly Using Claude Data
Anthropic says China-based labs including Alibaba and Moonshot AI covertly used millions of Claude exchanges to train rival AI models.
Anthropic has accused multiple China-based artificial intelligence companies, including tech giant Alibaba and startup Moonshot AI, of secretly harvesting millions of conversations with its Claude AI models to train their own competing systems, the company disclosed publicly. The alleged scheme represents one of the more brazen cases of unauthorized AI data extraction to surface from a major U.S. AI developer.
According to Anthropic, the activity was detected internally and constitutes a direct violation of the company's terms of service. The San Francisco-based AI safety firm did not specify precisely how many exchanges were compromised, but characterized the scale as involving millions of interactions — a volume that, if accurate, would represent a significant trove of training signal for any AI lab.
Read more Trust Stamp Broadens ID Dataweb Partnership Across Enterprise Base →
The accusations arrive at a fraught moment in the global AI race, with U.S. and Chinese technology companies locked in fierce competition to develop the most capable large language models. Anthropic's Claude is among the leading frontier models in the U.S., and any unauthorized access to its outputs could theoretically accelerate rivals' development timelines without requiring the same investment in data infrastructure.
The incident underscores growing concerns in the AI industry about so-called "model distillation" abuse — a technique where outputs from a powerful model are used to cheaply fine-tune a weaker one. Industry observers note that while distillation is a legitimate research practice when done within sanctioned boundaries, covert large-scale extraction raises serious legal and ethical questions that regulators have yet to fully address.
Continue reading at US Top News and Analysis.
