Chinese researchers used models built by OpenAI and Anthropic to help train a domestic Chinese AI system, according to a report published by NDTV, which cites reporting on the practice.
The detail at the center of the story is a familiar one in AI circles: rather than building a capable model entirely from scratch, developers can lean on an existing, more advanced system to help shape a new one — using its outputs as training material. NDTV's report frames the case as Chinese researchers drawing on American commercial models to advance work on a homegrown alternative.
Beyond that, the publicly available details are thin. The source item does not specify which particular OpenAI or Anthropic models were involved, which Chinese institution or company was responsible, when the work took place, or how the researchers obtained access. Neither OpenAI nor Anthropic is quoted in the source material, and no response from either company is included.
The context that makes this newsworthy is the broader U.S.–China technology competition. American AI companies generally prohibit customers from using their systems' outputs to build competing models, and Washington has spent years restricting China's access to advanced chips and AI capabilities. A case in which leading U.S. models allegedly served as raw material for a Chinese system would cut against both the commercial terms of service and the policy goal.
It also raises a harder question that export controls do not address well: hardware can be tracked at a border, but model outputs are just text, and they travel easily.
Readers should treat the specifics with appropriate caution — this is a single report summarized in a news aggregator item, and the underlying evidence, methodology, and company responses are not laid out in the source available here.
Why it matters: if capable AI systems can be effectively copied through their own outputs, the technological lead the U.S. has tried to protect through chip restrictions may be far more porous than policymakers assume.