People’s Daily warns of countermeasures over US AI theft claims

Mozilla published version 1.1 of its State of Open Source AI report on Monday, finding that the best Chinese open-weight AI models now lag leading U.S. closed models by roughly four months on capability benchmarks — and cost a fraction of the price. The update lands amid escalating tensions over allegations that Chinese firms trained their models using data extracted from American systems.

The Narrowing Gap

The report, built on a survey of about 1,400 developers and data from OpenRouter, Artificial Analysis, and the research nonprofit METR, found that the top open model scored within three points of the leading closed model on the Artificial Analysis Intelligence Index at 60% of the price. Mozilla's analysis of METR task-horizon data placed the open-closed gap at about 4.4 months, consistent with Epoch AI's own four-month estimate.tomshardware

On OpenRouter, a marketplace that routes developer API traffic across hundreds of models, eight of the top ten models by August token volume were open-weight, with seven of those built by Chinese companies. Yet closed providers still captured 96% of model-layer revenue on the platform from May through September 2025, according to the Linux Foundation. Mozilla CTO Raffi Krikorian told Ars Technica that the decision to pay for closed models is "workload-specific rather than organization-specific".tomshardware

The report cautioned that the four-month figure and cost comparisons are measured API-to-API at list price. Running models locally widens the gap: the best open model that fits on a single server scored 52.6, and the best on a single GPU scored 40 — drops of 10 and 23 points, respectively. Mozilla's data also stops at September 1; updated benchmarks published since then show a wider spread on some leaderboards.tomshardware

Distillation Dispute

The report arrives against a backdrop of U.S.-China friction over AI intellectual property. On September 8, the NSA, CISA, and FBI issued a joint advisory accusing six Chinese AI companies — including Moonshot AI, DeepSeek, Alibaba ↘ 1.17%, MiniMax, StepFun, and Z.AI — of conducting "industrial-scale" distillation campaigns against American frontier models. The advisory alleged that Moonshot extracted data from Anthropic's Claude Fable 5 to train its Kimi K3 model. Anthropic claimed that Chinese labs generated over 16 million exchanges with Claude through roughly 24,000 fraudulent accounts.cisa+2

Mozilla's report noted the allegation but described it as "asserted, and unshown".tomshardware

Beijing Pushes Back

China's People's Daily rejected the U.S. claims in a commentary published on Wednesday, calling them "without factual or legal basis" and accusing Washington of "politicising" what it described as a normal technical practice. The state newspaper warned that Beijing would take "countermeasures" if the allegations were used to contain Chinese firms. China's Commerce Ministry similarly called distillation "a neutral technical approach used by model developers around the world, including US companies".scmp+2

The exchange comes a week before an expected meeting between President Xi Jinping and President Donald Trump on September 24, where AI governance is expected to be a point of discussion.scmp