Anthropic uncovers unauthorized Claude usage by Chinese AI firms

Anthropic announced that it has identified illicit activity by two China‑based artificial‑intelligence labs, finding that they tapped into its Claude language model to accelerate their own projects. The company’s security team traced a pattern of usage that far exceeded normal API traffic, indicating systematic extraction of the model’s capabilities.

CNBC reported that the labs involved are the research division of Alibaba Group and the startup Moonshot AI. Both entities allegedly made use of millions of Claude exchanges, effectively harvesting the model’s responses to train or fine‑tune their own systems. Anthropic said the activity was “unauthorized” and that it has since implemented stricter access controls to prevent further exploitation.

In response, Anthropic has blocked the offending accounts and is working with its legal team to assess possible remedies. The firm emphasized that its terms of service prohibit any commercial or research use of Claude without a proper licensing agreement, a rule that the Chinese labs appear to have ignored.

The episode highlights a growing tension in the global AI arena. Since the launch of large‑language models, companies have increasingly packaged them as licensed services, charging developers for API calls. Unauthorized scraping not only undermines revenue streams but also raises concerns about intellectual‑property theft and national security, especially as the United States tightens export‑control rules on advanced AI technologies. Chinese firms have historically pursued rapid advancement in AI, sometimes by reverse‑engineering foreign models, a practice that has drawn scrutiny from both industry watchdogs and government regulators.

Analysts say the fallout could affect the competitive landscape. If Chinese developers are forced to build their own models from scratch, progress may slow, while firms like Anthropic could see a boost in market confidence after demonstrating robust detection capabilities. Conversely, the incident may prompt other AI providers to reinforce their security measures, potentially reshaping how cross‑border AI collaborations are managed.

The situation remains fluid as Anthropic evaluates the full scope of the breach and considers further legal steps. For now, the company’s swift action underscores the importance of safeguarding AI assets in an increasingly contested technological battlefield.