Reuters reported that China is weighing curbs on foreign access to its most advanced AI models. The Commerce Ministry is said to have met tech firms such as Alibaba and ByteDance about possible export controls, and even about treating leaks of core AI technology as a national-security risk. Chinese models are far cheaper and increasingly popular abroad, so any curbs could sharply raise costs for foreign firms that rely on them.
人工智能 (rén gōng zhì néng) artificial intelligence (AI)
模型 (mó xíng) a model; an AI model or system
出口管制 (chū kǒu guǎn zhì) export controls
开源 (kāi yuán) open source; free for anyone to download and modify
先进 (xiān jìn) advanced; cutting-edge
性能 (xìng néng) performance; how well something works
核心 (hé xīn) the core; the central, most important part
分级 (fēn jí) to tier; to classify by level
平台 (píng tái) a platform
泄露 (xiè lù) to leak; to divulge (secrets)
限制 (xiàn zhì) to restrict; to curb
收紧 (shōu jǐn) to tighten (rules or policy)
人工智能 (rén gōng zhì néng) artificial intelligence (AI)
模型 (mó xíng) a model; an AI model or system
出口管制 (chū kǒu guǎn zhì) export controls
開源 (kāi yuán) open source; free for anyone to download and modify
先進 (xiān jìn) advanced; cutting-edge
性能 (xìng néng) performance; how well something works
核心 (hé xīn) the core; the central, most important part
分級 (fēn jí) to tier; to classify by level
平台 (píng tái) a platform
洩露 (xiè lù) to leak; to divulge (secrets)
限制 (xiàn zhì) to restrict; to curb
收緊 (shōu jǐn) to tighten (rules or policy)
7月7日,路透社独家报道,中国正在考虑限制外国使用中国最先进的人工智能(AI)模型。
据知情人士透露,过去一个月,中国商务部与阿里巴巴、字节跳动等科技企业举行了多次会议,讨论是否对最先进的AI模型实施出口管制。会议还讨论了是否将泄露核心AI技术认定为危害国家安全的行为。
截至目前,相关企业没有对此作出回应,中国官方也尚未公开证实这一消息。不过,中国最高人民法院的一份刊物此前曾刊登专家建议,提出对AI模型进行分级管理。
中国的AI模型价格低、性能强,在全球得到了广泛使用。据CNBC报道,在美国AI模型平台OpenRouter上,使用中国模型的成本通常比美国同类产品低60%到90%,中国模型的使用量已连续数月超过三成。
为了节省成本,一些美国公司把日常任务交给便宜的中国模型,只在处理复杂问题时才使用昂贵的美国模型。据报道,有公司改用中国模型后,一年节省了数百万美元。
分析人士指出,一旦中国正式限制出口,依赖中国模型的外国企业将面临成本明显上升的压力。不过,相关措施仍处于讨论阶段。由于已经发布的开源模型可以自由下载、无法收回,新规定可能只针对未来的新模型。
值得注意的是,美国近期同样收紧了对本国最先进AI模型的出口限制。由此可以看出,先进的AI技术正在被中美两国当作重要的资源加以保护。
7月7日,路透社獨家報道,中國正在考慮限制外國使用中國最先進的人工智能(AI)模型。
據知情人士透露,過去一個月,中國商務部與阿里巴巴、字節跳動等科技企業舉行了多次會議,討論是否對最先進的AI模型實施出口管制。會議還討論了是否將洩露核心AI技術認定為危害國家安全的行為。
截至目前,相關企業沒有對此作出回應,中國官方也尚未公開證實這一消息。不過,中國最高人民法院的一份刊物此前曾刊登專家建議,提出對AI模型進行分級管理。
中國的AI模型價格低、性能強,在全球得到了廣泛使用。據CNBC報道,在美國AI模型平台OpenRouter上,使用中國模型的成本通常比美國同類產品低60%到90%,中國模型的使用量已連續數月超過三成。
為了節省成本,一些美國公司把日常任務交給便宜的中國模型,只在處理複雜問題時才使用昂貴的美國模型。據報道,有公司改用中國模型後,一年節省了數百萬美元。
分析人士指出,一旦中國正式限制出口,依賴中國模型的外國企業將面臨成本明顯上升的壓力。不過,相關措施仍處於討論階段。由於已經發布的開源模型可以自由下載、無法收回,新規定可能只針對未來的新模型。
值得注意的是,美國近期同樣收緊了對本國最先進AI模型的出口限制。由此可以看出,先進的AI技術正在被中美兩國當作重要的資源加以保護。
On July 7, Reuters reported exclusively that China is considering restricting foreign use of its most advanced artificial intelligence (AI) models.
According to people familiar with the matter, over the past month, China's Commerce Ministry and firms such as Alibaba, ByteDance and other tech companies have held several meetings to discuss whether to impose export controls on the most advanced AI models. They also discussed whether leaking core AI technology should be classified as endangering national security.
So far, the companies involved have not responded, and Chinese officials have not publicly confirmed the report. However, a publication of China's Supreme People's Court had earlier carried an expert proposal suggesting tiered management of AI models.
China's AI models are cheap and high-performing, and are widely used around the world. According to CNBC, on the U.S. AI-model platform OpenRouter, using Chinese models typically costs 60-90% less than comparable U.S. products, and Chinese models' usage has topped 30% for several months running.
To save on costs, some U.S. firms hand routine tasks to cheaper Chinese models, using pricier U.S. models only for complex problems. One company reportedly switched to Chinese models and saved millions of dollars a year.
Analysts note that once China formally restricts exports, foreign firms that rely on Chinese models will face pressure from sharply rising costs. Still, the measures remain at the discussion stage. Because already-released open-source models can be freely downloaded and cannot be recalled, any new rules may target only future models.
Notably, the U.S. has recently likewise tightened export limits on its own most advanced AI models. This shows that advanced AI technology is being treated by both China and the U.S. as an important resource to protect.