Jason Hsu, Senior Fellow at the Hudson Institute

The video discusses the issue of China's political strategy to steal US frontier models. Regarding this part, I think this is of great concern to US commercial entities and laboratories working on frontier models, and they may request President Trump to express dissatisfaction directly to Xi Jinping on this matter.

The second part is that regarding cybersecurity, cyberattacks, and cyberdefense, the two sides may establish a so-called guardrail mechanism. That is, is it possible that when an AI agent crosses a red line, there must be firebreaks—a switch-on and switch-off mechanism—to prevent the AI agent from taking over the entire system.

Each iteration of trusted technology allows the two sides to form a so-called trusted mechanism for certain sensitive industries, such as national defense cybersecurity or even finance, and then develop their model within a controllable range.

Through the Taiwan-US cooperation mechanism, an effective scope is delineated so that relevant classified data can be trained within it, and after this training, it can be used within a controllable range. Through this mechanism, a so-called trusted AI supply chain is formed, which I think is something both sides can consider. Then, through this AI supply chain mechanism, it can also be promoted to applications for other similar allies.

Video Link:

Full Text: