At the AI safety governance sub-forum of China's 2026 National Cybersecurity Week in Jinan on September 15, three groups of datasets were released. Under the Cyberspace Administration of China's guidance, the Cyberspace Security Association of China, together with the National Computer Network Emergency Response Center (CNCERT) and others, published Chinese Internet Corpus 4.0, a roughly 120GB collection of high-quality, verified Chinese-language data and the fourth in the series. CNCERT and partners also released Cybersecurity Dataset 1.0, covering five sub-datasets such as a security capability benchmark, APT attack-defense data, alert data, abnormal traffic, and vulnerability analysis, plus an AIGC Detection Dataset of 100,000 text and 100,000 image samples. The datasets are available through the Chinese Internet Corpus Resource Platform.
Our takeThis is supply-chain policy in data form: instead of only regulating outputs, China is building public, vetted training inputs. A 120GB curated corpus and an AIGC-detection set of 200,000 samples are raw material for both stronger domestic models and tougher content-labeling enforcement. The practical question is access, not scarcity: a platform gated by registration and vetting keeps the data credible, but it also keeps it onshore.