The Trump administration is accusing China of carrying out industrial-scale theft of intellectual property from American AI firms, even as Washington has taken a permissive stance toward the use of copyrighted material to train artificial intelligence systems. According to the Financial Times, the warning was set out in a memo from Michael Kratsios, the White House Office of Science and Technology Policy director, which was circulated across government departments and asked officials to alert AI companies to any foreign attempts to reach sensitive information.

The dispute centres on a practice known as distillation, in which the output of a larger model is used to train a smaller one so it can imitate performance at lower cost. Anthropic has already accused several China-based AI laboratories of using that method, while OpenAI has alleged that DeepSeek relied on distillation to build its open-source model and was effectively trying to free-ride on the work of US companies.

That criticism lands in the middle of a broader and increasingly awkward debate over what counts as fair use in the AI era. Some companies have argued in court and in public that training on copyrighted material is transformative, yet recent reporting has shown how often model output can closely reproduce protected source material. At the same time, the legal status of AI-generated work remains unsettled: the US Copyright Office has said material created without human authorship is not eligible for protection.

The tension is sharpened by the way AI firms protect their own assets. The Washington Post reported that Anthropic has bought, scanned and destroyed millions of books for training purposes, a practice that has drawn scrutiny over how far copyright exceptions should stretch. Separately, the New York Times reported that when Anthropic’s Claude Code source leaked online, the company filed thousands of takedown requests, though a rewritten version on GitHub was treated as transformative and not covered by its claim.

Taken together, the episode shows a sector that wants broad latitude to ingest and learn from others’ material while demanding strong protection for its own code, model weights and outputs. That contradiction is now playing out not just in the courts, but in the geopolitics of AI, where Washington is increasingly casting Chinese rivals as both competitors and alleged thieves.

Source Reference Map

Inspired by headline at: [1]

Sources by paragraph:

Source: Noah Wire Services