# China’s AI Rollout Has No Off Switch

> China’s AI strategy lacks a kill switch, enabling rapid global diffusion without safety controls.
- Title: China’s AI Rollout Has No Off Switch · Foreign Policy
- Summary: China’s AI strategy lacks a kill switch, enabling rapid global diffusion without safety controls. China’s AI rollout has no off switch, allowing widespread…
- Keywords: china, ai safety, open-weight models, ai diffusion, kill switch, ai rollout, chinese ai models, ai adoption, ai agents, industrial ai, robotics, Rollout
- Source: Foreign Policy — https://foreignpolicy.com/2026/10/07/china-ai-race-artificial-intelligence-safety-models-kill-switch
- Author: Agathe Demarais
- Published: 2026-10-07T00:00:00+00:00
- Read time: 1 min
- Topics: ai, china, ai safety, open-weight models, ai diffusion, kill switch, ai rollout, chinese ai models, ai adoption, ai agents, industrial ai, robotics
## Historical Parallel
In 1957, Washington projected Soviet intercontinental missiles would reach 500 by 1960, but the count was wrong.
## AI Race Strategy
Beijing measures AI success through adoption and diffusion across its economy, not technological benchmarks.
## China’s AI Models
China’s best models, like Kimi K3, Qwen 3.8 Max, and GLM-5.3, are months behind U.S. frontier models.
## AI Plus Plan
China’s 2025 AI Plus action plan targets 70% AI agent penetration by 2027 and 90% by 2:30.
## Industrial Data
Beijing designates data as a factor of production to boost manufacturing through ai.
## Industrial Applications
By mid-2025, nearly half of China’s industrial firms deployed ai models, aiming for 1,000 agents by 2027.
## Embodied AI
China installed 354,000 industrial robots in 2025, producing 12,800 humanoid robots, 90% of global total.
## Open-Weight Models
Chinese ai models are open-weight, allowing download and fine-tuning without server restrictions.
## Self-Reinforcing Cycle
Chinese models become industry standards as developers adapt them for specific use cases.
## Safety Implications
U.S. safety debates focus on kill switches for closed models, but open-weight models lack recall mechanisms.
## GLM-5.3 Release
Zhipu delayed publishing GLM-5.3 weights for two weeks due to cybercapabilities concerns.
## Model Adaptations
Developers created 80 variants of GLM-5.3, including security-research tools that remove safety guardrails.
## U.S. Adoption
Nvidia published a compressed version of GLM-5.3 to optimize disk space and memory use.
## Invisible AI
China’s ai advances are invisible due to widespread diffusion, making them hard to track or stop.
## Key takeaway

China’s AI rollout has no off switch, allowing widespread adoption without safety mechanisms.