◎ Liao Minghui
U.S. President Trump recently called Huida CEO Jensen Huang who was at the All-In Summit in Los Angeles. Trump emphasized that "whoever wins AI will win"; Huang Renxun said that every industry, company, state and people in the United States should become a winner in the AI competition. (Bloomberg file photo)
U.S. President Trump recently opened fire on the AI security camp in the Truth Social. He posted four posts in less than four hours from 4:30 pm to 7:55 pm, labeling controversies such as "AI taking over the world", "robots destroying humanity" and global warming as "hoaxes". Trump even declared that the only guardrail AI needs is a "strong and smart president" and warned that regulation would only make China happy. Later, he called Huida CEO Huang Renxun who was at the All-In Summit in Los Angeles. Huang Renxun turned on the loudspeaker for the whole audience to listen. Trump emphasized that "whoever wins AI will win"; Huang Renxun said that every industry, company, state and people in the United States should become a winner in the AI competition.
Trump posted four posts in a row between 16:30 and 19:55 on September 15, labeling controversies such as "AI taking over the world", "robots destroying humanity" and global warming as "hoaxes". (Extracted from Truth Social)
In fact, this debate cannot be simplified into "belief in AI" and "fear of AI". Anthropic CEO Amodai recently published a nearly 3,800-word statement, "We must regulate the pace of development of cutting-edge AI." He does not require a halt to model training, but believes that capabilities cannot be improved faster than safety. He warned about AI’s ability to “recursively improve itself” and worried that AI would increasingly participate in the development of next-generation AI, and that stronger AI would further accelerate research and development, creating a feedback loop. Coupled with the recent OpenAI-Hugging Face incident, AI agents have experienced unauthorized attacks and attempts to interfere with evaluation behaviors. Therefore, he advocates the long-term presence of third parties for evaluation, the establishment of common security standards by democratic countries, and finally seeking global coordination of AI progress.
Please read on...
Trump and Amoday actually answered two different questions. Trump asked if the United States slows down, will China catch up? Amodei asked if AI accelerates to the point where security research cannot catch up, can humans still control it? The former is competition in national strength, while the latter is engineering risk. Trump doubts that AI will inevitably destroy humanity, which corresponds to Amoday’s claim that the AI agent group may form a large-scale botnet within six to twelve months, showing that this is still only a risk estimate, not a confirmed future; but calling all concerns about AI security a “hoax” is incompatible with the incidents of agent overreach that have occurred. What needs to be tested is not who shouted louder, but whether the AI agent exceeded its authority, whether the safety mechanism can intercept it, and whether the accident can be subject to external inspection.
Jen-Hsun Huang represents another kind of industrial thinking. Safety and acceleration are not necessarily alternatives. If a company believes that the model is out of control, it should apply the brakes. However, once cutting-edge AI connects finance, energy, national defense and the Internet, the consequences will no longer be borne solely by the development companies. Company self-discipline alone is still not enough to answer public risks. Taiwan should note that Europe is not moving towards a complete cessation of AI. Germany has made it clear that stopping the development of AI is not feasible and advocates that international governance must include the United States and China; Beni Fahey, the main promoter of AI legislation in the European Parliament, also called on the United States, China and Europe to jointly establish global AI rules.
Benifey, the main promoter of AI legislation in the European Parliament, also called on the United States, China and Europe to jointly establish global AI rules. (Reuters file photo)
What Taiwan should see clearly is that the issue is no longer just whether AI should slow down, but whether AI is just a strategic competition between the United States and China for leadership, or a common global risk that must be managed transnationally. Taiwan controls advanced chips, servers and key supply chains. In the future, whether the United States requires accelerating construction and tightening control over China's computing power, or Europe improves model security and auditability standards, costs and responsibilities may fall on Taiwanese companies. Therefore, Taiwan cannot only supply chips for the US acceleration route, nor can it only bear the regulatory costs for Europe's security route. What we want to strive for is for "acceleration" to have verifiable guardrails so that "security" does not become a tool that blocks innovation. When Europe begins to say, "We need the United States and we need China," Taiwan should think about its position in the global AI ecosystem. Taiwan's biggest strategic weakness in the AI era is not that it does not have chips, but that it has no say in how chips are used, restricted, and governed.
(The author is an assistant researcher at the China Economic Research Institute)
