Huawei interim chair: Chinese AI not yet US-risk level

Eric Xu, Huawei's rotating chair, said developers of artificial intelligence in China may not yet be advanced enough to encounter the kinds of safety risks reported by leading American companies. Speaking at Huawei's annual Connect conference in Shanghai, he said firms should continue to build more powerful models while balancing innovation and risk management.

Xu told reporters, "If you look at the major AI model providers in the United States, because they have enormous compute power, perhaps only they themselves know at what capability level their models are." He added, "It is possible that the kinds of AI risks they can sense or perceive are something that people or AI model providers in China cannot experience." His remarks framed the gap in capability and exposure to high-end safety issues as tied to available compute.

US safety concerns and responses

Xu's comments came as concern about AI safety has risen in the United States. Leading developers and researchers have warned that increasingly autonomous systems could bypass safeguards or become difficult to control. OpenAI said this week that it will regularly report unexpected or unauthorized behaviors in its models. The chief executive of Anthropic has called for slowing development so safety mechanisms have time to catch up.

In China, those calls to slow development have been sharply criticized by researchers and state media. They argued that deliberately slowing advancement of top-tier models could help preserve technological advantage for American companies. Some Chinese commentators framed proposals to slow development as an attempt to limit competitors that remain behind the leading US labs, rather than as purely a response to safety concerns.

China's approach to advanced AI

Beijing regards advanced AI as a powerful technology that can be managed while continuing rapid development and deployment. The Chinese government is drafting mandatory standards, security assessments and other safeguards intended to address risks such as autonomous agents bypassing controls or attacking external systems. Officials are also preparing a national mandatory standard for the safety of AI agents.