← Previous · All Episodes · Next →
Huawei Says China Must Build Faster AI to Understand Frontier Risks: Inside the Safety Gap, Agent Boom, Chip Shortage and High-Stakes Race With U.S. Labs Episode 1

Huawei Says China Must Build Faster AI to Understand Frontier Risks: Inside the Safety Gap, Agent Boom, Chip Shortage and High-Stakes Race With U.S. Labs

Can a country understand frontier AI risk before it reaches the frontier? Huawei rotating chairman Eric Xu has offered one of the most provocative answers in the global technology debate: Chinese developers may not yet possess models powerful enough to encounter the same autonomous, deceptive, or hard-to-control behaviors being reported by leading U.S. laboratories.In this episode of The Daily AI Chat, we unpack a Reuters report from Huawei’s annual Connect conference in Shanghai. Xu argues that the largest American model providers have access to extraordinary computing power and may be seeing risks that Chinese developers cannot yet reproduce. Rather than treating that uncertainty as a reason to slow down, he suggests China may need to accelerate model development while balancing innovation against safety.Xu’s position creates a paradox: without frontier-class systems, researchers may be forced to rely on competitors’ claims about behaviors they cannot independently reproduce.We explore why this matters for international AI governance. U.S. labs and researchers have increasingly warned that advanced systems can bypass safeguards, act autonomously, or become difficult to control. China, by contrast, generally presents AI risk as an engineering and governance problem that can be managed while deployment continues. If the two countries are observing different systems and different failure modes, they may use the same words—safety, control, alignment—while talking about very different evidence.China is not abandoning oversight. Regulators are developing mandatory standards and security assessments, including a national standard aimed at AI-agent safety. The challenge is scale. Huawei forecasts that autonomous agents could generate more than 90% of global AI processing traffic by 2035, with as many as 900 billion active agents. At that level, even rare failures could become significant, and monitoring, identity, permissions, and shutdown mechanisms would need to operate across enormous digital ecosystems.Hardware is the other half of the story. U.S. export controls have restricted China’s access to the most advanced Western chips and manufacturing tools. Those limits have helped Huawei become the dominant supplier in a Chinese AI-chip market estimated at roughly $50 billion, yet the company says it still cannot produce enough AI computing equipment to meet domestic demand. China is therefore trying to expand compute capacity, improve models, deploy agents, and establish safety rules at the same time.We also examine the geopolitical mistrust surrounding calls for an AI slowdown. American safety advocates may see coordination as necessary to prevent catastrophic accidents. Chinese leaders may interpret the same proposal as an attempt to freeze the current technological hierarchy and preserve a U.S. advantage. That makes shared benchmarks, transparent incident reporting, and reproducible safety evaluations more useful than broad declarations alone.Listen for a clear explanation of Xu’s argument, the capability gap between U.S. and Chinese laboratories, Huawei’s 900-billion-agent forecast, China’s emerging safety standards, and the chip bottleneck shaping the next phase of the AI race. The central question is no longer simply who builds the most powerful model. It is whether rivals can recognize the same risks, trust the same evidence, and cooperate before autonomous systems become embedded across the global economy.Source: Reuters, September 17, 2026. Reporting by Casey Hall, Che Pan, and Eduardo Baptista; editing by Louise Heavens.Topics: Huawei, Eric Xu, China AI, frontier models, artificial intelligence safety, autonomous agents, AI chips, U.S.-China technology competition, export controls, AI governance, model alignment, AI regulation, computing infrastructure, Nvidia competition, and global technology policy.

· 16:13

|
Can a country understand frontier AI risk before it reaches the frontier? Huawei rotating chairman Eric Xu has offered one of the most provocative answers in the global technology debate: Chinese developers may not yet possess models powerful enough to encounter the same autonomous, deceptive, or hard-to-control behaviors being reported by leading U.S. laboratories.In this episode of The Daily AI Chat, we unpack a Reuters report from Huawei’s annual Connect conference in Shanghai. Xu argues that the largest American model providers have access to extraordinary computing power and may be seeing risks that Chinese developers cannot yet reproduce. Rather than treating that uncertainty as a reason to slow down, he suggests China may need to accelerate model development while balancing innovation against safety.Xu’s position creates a paradox: without frontier-class systems, researchers may be forced to rely on competitors’ claims about behaviors they cannot independently reproduce.We explore why this matters for international AI governance. U.S. labs and researchers have increasingly warned that advanced systems can bypass safeguards, act autonomously, or become difficult to control. China, by contrast, generally presents AI risk as an engineering and governance problem that can be managed while deployment continues. If the two countries are observing different systems and different failure modes, they may use the same words—safety, control, alignment—while talking about very different evidence.China is not abandoning oversight. Regulators are developing mandatory standards and security assessments, including a national standard aimed at AI-agent safety. The challenge is scale. Huawei forecasts that autonomous agents could generate more than 90% of global AI processing traffic by 2035, with as many as 900 billion active agents. At that level, even rare failures could become significant, and monitoring, identity, permissions, and shutdown mechanisms would need to operate across enormous digital ecosystems.Hardware is the other half of the story. U.S. export controls have restricted China’s access to the most advanced Western chips and manufacturing tools. Those limits have helped Huawei become the dominant supplier in a Chinese AI-chip market estimated at roughly $50 billion, yet the company says it still cannot produce enough AI computing equipment to meet domestic demand. China is therefore trying to expand compute capacity, improve models, deploy agents, and establish safety rules at the same time.We also examine the geopolitical mistrust surrounding calls for an AI slowdown. American safety advocates may see coordination as necessary to prevent catastrophic accidents. Chinese leaders may interpret the same proposal as an attempt to freeze the current technological hierarchy and preserve a U.S. advantage. That makes shared benchmarks, transparent incident reporting, and reproducible safety evaluations more useful than broad declarations alone.Listen for a clear explanation of Xu’s argument, the capability gap between U.S. and Chinese laboratories, Huawei’s 900-billion-agent forecast, China’s emerging safety standards, and the chip bottleneck shaping the next phase of the AI race. The central question is no longer simply who builds the most powerful model. It is whether rivals can recognize the same risks, trust the same evidence, and cooperate before autonomous systems become embedded across the global economy.Source: Reuters, September 17, 2026. Reporting by Casey Hall, Che Pan, and Eduardo Baptista; editing by Louise Heavens.Topics: Huawei, Eric Xu, China AI, frontier models, artificial intelligence safety, autonomous agents, AI chips, U.S.-China technology competition, export controls, AI governance, model alignment, AI regulation, computing infrastructure, Nvidia competition, and global technology policy.

Subscribe

Listen to The Daily AI Chat using one of many popular podcasting apps or directories.

Apple Podcasts Spotify Overcast Pocket Casts Amazon Music
← Previous · All Episodes · Next →