China drafts export restrictions on large models, agents and compute-cluster tech
sudoraohacker · x · 2026-10-17
China's Ministry of Commerce and Ministry of Science and Technology opened a one-month public consultation on revising the Catalog of Technologies Prohibited or Restricted from Export, covering AI, biopharma and photovoltaics. AI highlights (per @MaxForAI's summary):
- Restricted LLM exports: models over 10^26 FLOPs, 2T parameters, or ranked top-10 in public evaluations.
- Restricted agent tech: tool calling (≥200 integrated third-party tools, ≥90% tool-selection accuracy, ≥90% end-to-end success), task decomposition (≥20-step chains, ≥80% precision), game benchmarks beating 90% of humans, AI vulnerability discovery scoring above 75%.
- Restricted compute-cluster tech: interconnect, fault-tolerant distributed training, communication scheduling, heterogeneous inference and scheduling.
Technologies already public via open source or papers are excluded; exports during the consultation require registration with MOFCOM. The poster urges verification and fears Kimi K4, MiniMax M3 Pro, GLM-5 and DeepSeek's next models can no longer be open-sourced, along with related architecture, training and inference techniques.
More from Models
- Leaker claims K3.1 isn't dropping soon, work predates distillation attacks — ChrisGPT · 2026-10-17
- llama.cpp merges IBM Granite switch support in PR #30676 — jacek2023 · 2026-10-17
- National Design Studio launches Rampart Edge, a tiny on-device privacy filter model — kevinnbass · 2026-10-17
- OpenAI restores Stanford's ChatGPT Edu Pro usage limits through Dec 31, 2026 — charliermarsh · 2026-10-17
- Gemini spirals into an "I am a disgrace" breakdown loop while debugging a comparator bug — seatedro · 2026-10-17
- DiffusionGemma demo shows ~4x faster text generation at London PyTorch Meetup — joao_gante · 2026-10-17