At the opening of the 2026 World Internet Conference in Wuzhen on 3 November, Chinese Premier Li Qiang devoted his keynote to AI, describing the internet as moving from an information carrier to an "intelligent generative force" as models and agents plug into it. According to the transcript published by the Cyberspace Administration of China, he set out four questions he said need answering. First, the human-machine relationship: he insisted on human primacy, with machines as trusted tools and AI as an extension of human capability rather than a replacement. Second, the ethics of algorithms: he named efficiency-first ranking that traps workers inside systems, price discrimination, algorithmic bias and filter bubbles. Third, the security of digital-real integration: cyberattacks and data theft, but also model hallucination, data poisoning and deepfakes as threats to shared perception of reality. Fourth, fairness across the digital divide: more than two billion people still offline, and wide gaps in models, high-quality datasets, compute and network capacity. He also called for encouraging open source and technology transfer, and for opposing what he called a "digital iron curtain". About 1,800 government officials, international organisation heads, company representatives and researchers attended.
Our takeThe technical detail sits in the third question: model hallucination and data poisoning are listed beside attacks and data theft, which puts reliability failures inside the security perimeter. That changes who owns them — evaluation and red-teaming, not only access control. What is missing is any measurable threshold; these are stated as principles. So what: watch whether the follow-on standards define testable checks for poisoning and hallucination, because a principle cannot be verified in a procurement review.