Anthropic Says Chinese AI Labs Used Claude Data to Train Rivals
Anthropic detected unauthorized use of its Claude AI by Chinese labs including Alibaba and Moonshot AI to improve competing models.
Anthropic has accused several China-based artificial intelligence companies, including tech giant Alibaba and startup Moonshot AI, of secretly leveraging millions of exchanges with its Claude AI models to train their own competing systems — a disclosure that raises fresh alarms about intellectual property theft in the global AI race.
The San Francisco-based AI safety company said it identified the unauthorized activity after detecting suspicious usage patterns linked to the Chinese labs. Rather than building their own training datasets from scratch, these entities allegedly routed large volumes of queries through Claude to extract outputs that could then be fed into rival model development pipelines — a technique sometimes called model distillation or data harvesting.
Read more Apple iPhone Duo Enters China's Foldable Market Amid Price Scrutiny →
The revelation underscores a growing vulnerability facing American AI firms: open API access, designed to democratize cutting-edge technology, can simultaneously expose proprietary model behavior to competitors willing to exploit it at scale. Anthropic's terms of service prohibit using Claude outputs to train other AI models, making the alleged conduct a clear terms violation and potentially a legal matter.
The incident adds to mounting tensions between U.S. and Chinese AI ecosystems, where competition has intensified following the emergence of low-cost Chinese models that have already rattled American technology stocks. For Anthropic, a company that has positioned itself as a safety-first AI developer backed by billions in investment, the breach represents both a reputational and competitive challenge that it chose to disclose publicly rather than handle quietly.
Continue reading at US Top News and Analysis
