China's Ministry of Commerce and Ministry of Science and Technology published Announcement No. 48 on Oct 16, proposing to add 40 technologies (23 in AI, plus biomedicine and solar) to the catalogue of technologies banned or restricted from export. Comments are open until Nov 15. Pre-export registration on MOFCOM's platform applies immediately, including exports already under way. Technology already public through open source or published papers is excluded. The draft annex's banned list includes AI-driven malicious code generation and mutation usable against critical infrastructure, defence or core government systems that reaches 60% attack success against multi-engine detectors, 85% against a single mainstream antivirus or 90% against ML models. It also covers sandbox, debugger and emulator detection or sandbox escape, rewriting at least 50% of code structure per run, and batch generation of at least 20 evasive variants. The restricted list adds AI-assisted vulnerability discovery that meets any of several bars. Examples are at least 75% on authoritative exploitability benchmarks, at least 50% end-to-end exploit-chain success against mainstream OSes or browsers, a 10x efficiency gain over senior experts, chaining two or more low or medium bugs into a working PoC, or finding five or more new high-severity flaws in a major project in one run. Agents built for critical infrastructure, defence or government use at the 'high-performance' level are banned. Large models trained on China's 'core data' are banned too.