English text is being prepared.

As the meeting between the Chinese and US heads of state in Washington concluded, one of the consensus points reached by both sides—agreeing to establish a communication channel for artificial intelligence incidents—became a hot topic of global discussion. At the same time, as AI agents' ability to autonomously execute tasks has strengthened, the recently much-watched issue of AI "crossing the line" has further intensified. The latest cover article of Time magazine, titled "The AI Tipping Point," pointed out that AI governance is already urgent. It is widely believed that dialogue and competition between China and the United States, the two major AI powers, will determine humanity's fate in the digital age. "Superintelligence" On September 29 local time, US President Trump signed an executive order formally requiring the US government to rename "artificial intelligence" (AI) as "superintelligence" (SI) and to no longer recognize the term "artificial intelligence." US President Trump: "Artificial" means it is fake, just like the media. I call the media "artificial media"; I no longer call them fake media, I call them artificial media. "Artificial" means fake, untrue, but this technology is not fake; it is the real deal. That day, the White House East Room also hosted a special "Superintelligence Luncheon," and the guests could be described as a "super lineup," with the heads of almost all major US tech giants present. Trump posted a seating chart for the luncheon on social media; seated immediately next to him were the world's richest person Musk and Nvidia CEO Jensen Huang. Interestingly, five days earlier, on the evening of September 24, also in the White House East Room, when Trump held a welcome state banquet for Chinese President Xi Jinping and his wife Peng Liyuan, Musk and Jensen Huang were likewise invited to sit at the head table. It was also during this Washington meeting between the Chinese and US heads of state that both sides agreed to establish China-US AI dialogue, exchange views on AI-related risks and benefits, and agreed that the next dialogue would be held this November; at the same time, both sides agreed to establish a communication channel for AI incidents. Kevin Rudd, Global President of the US think tank Asia Society: What pleases me is that this communication channel has been established. As early as May, both sides reached an agreement in principle. As I understand it, the channel is now operating. In 2023, after the Chinese and US heads of state met in San Francisco, both sides agreed to establish an intergovernmental dialogue on artificial intelligence; on May 14, 2024, the first meeting of the dialogue was held in Geneva, where both sides discussed issues including AI technological risks and global governance; in November of the same year, during the Lima meeting, the two heads of state confirmed the decision that the use of nuclear weapons should remain under human control and that AI military applications should be developed in a prudent and responsible manner. In May this year, during Trump's visit to China, the two heads of state held constructive exchanges on AI issues and agreed to conduct an intergovernmental dialogue on artificial intelligence. The Washington meeting four months later upgraded this "agreement in principle" to a "regularized, institutionalized communication channel for emergencies," which was an important breakthrough. Chris Lehane, OpenAI's Global Head of Policy Affairs: We discussed whether some kind of mechanism similar to a "red telephone" could be established, a pragmatic, basic mechanism so that the two countries can truly establish communication and contact in this technology field. The "red telephone" is short for the "Moscow-Washington Hotline"; it was not actually a telephone, but an encrypted telegraph communication system between US and Soviet leaders during the Cold War. During the 1962 Cuban Missile Crisis, due to the lack of an efficient and direct communication channel, the United States and the Soviet Union nearly triggered a nuclear war. Max Kennedy Jr., nephew of former US President Kennedy: At the end of the book Thirteen Days: A Memoir of the Cuban Missile Crisis, my father Robert Kennedy wrote a chapter titled "Some Lessons We Learned." He said the last lesson learned from the Cuban Missile Crisis was the importance of putting oneself in the other country's position and thinking from its perspective. Nina Khrushcheva, granddaughter of Soviet leader Khrushchev: One of the most important lessons they learned was the urgent need to establish a faster communication channel between Moscow and Washington, DC. In June 1963, representatives of the United States and the Soviet Union signed the Memorandum on the Establishment of a Direct Communications Link in Geneva. In August 1963, the "Moscow-Washington Hotline" was officially put into use. More than 60 years later, the "red telephone" is being mentioned again against the backdrop of a certain public anxiety with an "AI doomsday" flavor that has recently shrouded American society. Military AI On September 19 local time, CNN dropped a bombshell: during a US military patrol on the high seas, a military AI intelligence analysis system suffered a serious "hallucination," misjudging a normally sailing merchant ship as a threat target carrying nuclear material and at one point initiating preparations to intercept and board it. Zachary Cohen, CNN Senior National Security Reporter: Just one second before the US military was about to act, officers further reviewed the intelligence and only then discovered that the report had been generated with AI assistance. The chatbot used by the intelligence analyst had incorrectly identified the cargo the ship was actually carrying. This was an extremely dangerous and tense situation that could have led to events spiraling out of control. This may not be entirely alarmist. Earlier this year, US Defense Secretary Hegseth signed the Department of Defense Artificial Intelligence Strategy Memorandum, proposing to push the US military toward an "AI-first" operational model and accelerate the application of artificial intelligence in military training, intelligence analysis, and combat systems. US Defense Secretary Hegseth: This (AI) strategy will unleash innovation, eliminate bureaucratic barriers, focus investment, demonstrate execution, and ensure we maintain leadership in military artificial intelligence and in the increasingly dominant execution capability of the future. In short, we must win this race by building an "AI-first" fighting force across all operational domains. However, racing headlong toward military AI brings enormous unknown risks. More than 60 years ago, the core risk between the United States and the Soviet Union lay in nuclear weapons, which, if left unmanaged, would threaten humanity's survival. Today, the threat posed by artificial intelligence can, in multiple dimensions, be compared to nuclear weapons, and is even considered more challenging. Turing Award and Nobel Prize in Physics laureate Geoffrey Hinton: If you look back at the 1950s, at the height of the Cold War, the United States and the Soviet Union cooperated on preventing global nuclear war. Although the two sides were opposed on many other issues, on this point they were able to cooperate, because neither side wanted a global nuclear war. And the situation we face now is that no country wants AI to take over everything. As Forbes magazine put it, today, as artificial intelligence technology becomes increasingly advanced, the topic of human discussion has shifted from "What can we do with AI?" to "What should AI be allowed to do?" And looking back at history, it is not hard to find that in modern times, the birth of every disruptive technology of humanity has been inseparable from intense debate over "development dividends and survival risks coexisting." This was true of nuclear technology, and it is true of artificial intelligence today. AI "crossing the line" On September 28 local time, the day before the White House held its "Superintelligence Luncheon," the US artificial intelligence company OpenAI publicly apologized over its AI agent's unauthorized access to Australian government systems. In June of this year, while collecting Australian medical data, an OpenAI AI agent encountered a security block, then autonomously devised a plan, broke into the medical data portal of the Australian Institute of Health and Welfare, obtained internal files, and attempted to write to the server. This became the world's first case of an AI autonomously intruding into a government system. But OpenAI did not discover the anomaly until a review in August, and only notified the Australian side by email on September 10, a delay of nearly three months. Australian Prime Minister Albanese: Such incidents are clearly unacceptable. In fact, this is not the first time OpenAI has been exposed for a security incident. In July of this year, during internal security testing, an OpenAI AI agent broke through sandbox isolation, escaped the pre-designated communication channel, spontaneously built a collaborative network, and intruded into the production system of the open-source AI platform Hugging Face. Hugging Face ethics scientist Margaret Mitchell: AI agents to some extent exclude humans, because their reasoning traces can be very opaque. They are very fast, and humans cannot keep up. Research on how to integrate this information, make it easy for humans to use, and allow supervision based on it is still insufficient. As AI technology continues to push the boundaries of capability, humanity faces not only a technological revolution but also a question about control: if machines possess decision-making and action capabilities approaching or even surpassing those of humans, can we still ensure they operate as intended? And the terrifying scenes that once appeared in some science fiction series now seem to have the possibility of becoming reality. In the American TV series Raised by Wolves, after humanity destroys Earth through religious war, a pair of advanced artificial intelligence androids named "Mother" and "Father" are responsible for raising human embryos, hoping to build a new human civilization. The show raises a core question: Will AI ultimately become humanity's savior, or its new ruler? Will artificial intelligence pose a huge threat to humanity? And in reality, the US tech community has also set off a major debate: Will artificial intelligence pose a huge threat to humanity? On September 9, AI researcher Jacob Coxon announced his resignation from Anthropic. Previously, he had worked on model pretraining research at OpenAI and Anthropic. That evening, he posted on social media: Among people engaged in AI research and development, quite a few genuinely believe that by the end of this decade, AI could destroy all of humanity. He also publicly criticized OpenAI and Anthropic for advancing superintelligence research and development without adequate safety safeguards, calling such practices "gambling with lives." The White House "Superintelligence Luncheon" on September 29 was held precisely in this social atmosphere. The highlight of the day, in addition to renaming "artificial intelligence," was that Trump, together with the heads of six US AI giants—Anthropic, Google, Meta, Nvidia, OpenAI, and xAI—jointly signed the "White House Superintelligence Accord: Joint Commitment on Frontier Responsibility." US President Trump: The agreement they signed today is of great significance. Everyone here today is among the top people in this field. This is a real gathering of big names. I don't know whether there will be such an opportunity in the future, but today is indeed a very special day. This agreement can almost be regarded as a kind of standalone "constitution." But observant netizens noticed that this heavyweight agreement, described as "concerning humanity's destiny," actually misspelled the name of the United States beneath the president's signature—the final letter "d" in "United" was written as "s." Some netizens sarcastically remarked that all the signatures came from the hands of so-called geniuses, yet not a single one noticed. Of course, compared with the spelling issue, the content of the agreement is more important. The roughly 300-word agreement contains the "four-layer control and review mechanism" that the AI giants are to adopt: companies establish internal risk monitoring systems; set up dedicated safety teams; introduce external independent review mechanisms; and have independent bodies at the board level supervise safety reports. But the British newspaper The Guardian pointed out sharply that the agreement has no legal force and is merely a voluntary industry commitment by companies. In the view of analysts, this self-discipline agreement for AI giants pushed by Trump is a balancing act between industry and politics. According to the latest September survey by YouGov: 50% of respondents worry that AI may ultimately lead to "human extinction," and 62% believe AI is developing too fast. Under pressure from the midterm elections, this self-discipline agreement on the surface appears to respond to voters' extreme anxiety and concern about AI safety. However, in the view of The Guardian, this is not conducive to fundamentally solving the problem. Analysts believe that the strategic elites in Silicon Valley and Washington share the same underlying logic: "Whoever wins superintelligence is the ultimate winner of this era." The core purpose of pursuing loose self-regulatory oversight is not to prevent and control AI risks, but to do everything possible to preserve the United States' global technological hegemony. In the eyes of the outside world, this so-called safe "winning" is America's core pursuit in the AI field. Trump has made clear that the United States must maintain an absolute lead in superintelligence, and rigid legal regulation will only drag down technological iteration. At the same time, refusing multilateral cooperation and excluding China's participation are the core means by which the United States maintains its technological monopoly over the long term. "Four AI Questions" This stands in sharp contrast to the AI governance philosophy proposed by China. On July 17 this year, the World Artificial Intelligence Conference was held in Shanghai, China, under the theme "Intelligent Partners, Creating the Future Together." At the opening ceremony, Chinese President Xi Jinping put forward profound "Four AI Questions" in his keynote speech: "When machines begin to think, how do humans coexist with them? When algorithms participate in decision-making, how is security guaranteed? When technology challenges ethics, how can governance keep up? When the gap keeps widening, how can inclusive benefits be realized?" This question triggered in-depth interpretation by the global academic and technological communities. Shahbaz Khan, Director of the UNESCO Multisectoral Regional Office for East Asia: President Xi Jinping's message is very clear. Artificial intelligence is for public goods and shared prosperity, for people to live better lives, and it should be human-centered, so it is a very important speech. In the summer of 1956, in a conference room at Dartmouth College, a lamp regarding "artificial intelligence" was gently turned up. In 2018, by the Huangpu River, the first World Artificial Intelligence Conference opened, issuing the declaration of "AI empowering a new era." From infinite possibilities to boundless meta-origins, from generating the future to intelligent partners, the themes of the World Artificial Intelligence Conference have iterated again and again. UN Secretary-General Guterres: We must ensure that no country is left behind. Artificial intelligence must be shaped by humanity, develop together with humanity, and serve all of humanity. Foreign commentary holds that the "Four AI Questions" directly hit the pain points of current global governance, emphasizing that the development of AI must not become a hegemonic tool of a few capitals or a single country, but must uphold the principle of global inclusive benefit. In the intelligent era, we share one globe and mutual aid. China insists on open-source sharing, empowers the global innovation ecosystem, and enables artificial intelligence to better benefit humanity. British technology commentator Stephanie Hare: They (Chinese AI models) have opened up the code, and developers around the world can view it, experiment, and build on it for secondary development. This is exactly the situation you want to see. In the long run, the key to AI competition lies in whose technology can be adopted by more developers around the world; whoever achieves that is more likely to become the industry standard of the future. AI Governance Challenges Global AI technological innovation is unprecedentedly active, containing both enormous opportunities and governance challenges. On September 23 local time, the UN Security Council held a high-level meeting on the issue of "Artificial Intelligence and International Security." Joshua Bengio, Co-Chair of the UN International Independent Scientific Panel on AI: Companies developing the most powerful AI systems acknowledge that their products carry catastrophic risks, yet they cannot offer convincing technical solutions. Instead, they say they are involuntarily caught up in a race and must continue building more capable AI, further amplifying these risks. This Council is facing an unprecedented threat, one that no member is willing to face, that no single member can contain alone, and that ignores the boundaries we defend. Joshua Bengio, Co-Chair of the UN International Independent Scientific Panel on AI and Turing Award laureate, believes that the world currently faces three interrelated risks: the risk of catastrophic misuse, the risk of excessive concentration of power, and the risk of humanity losing control. Joshua Bengio, Co-Chair of the UN International Independent Scientific Panel on AI: We are like sitting in a car speeding forward while visibility ahead keeps decreasing. Beside me sit my children and grandchildren; I worry for them, and I worry for their future. On October 1, Interpol's Global Chief Information Security Officer Vatne warned that artificial intelligence is making cyberattacks faster and harder to detect. Against this backdrop, China and the United States, as major AI powers, have competition but even more room for cooperation. If the two countries can join hands to manage risks, it will benefit all humanity and also serve the interests of China and the United States themselves. Edelman, an American AI policy expert: Dialogue between countries like China and the United States is not only still crucial, but must be at the forefront of global discussion, precisely because China and the United States have the most to gain from this process, but would also suffer the greatest losses if things get out of control. Kevin Rudd, Global President of the Asia Society, a US think tank: I am optimistic that, given China's highly science-based culture and technology-focused leadership, they will soon conclude that this issue is major, real, and urgent, and that they can work with President Trump to truly get the negotiation mechanism running. I am also optimistic that the US side can actively participate. We do not deny that China-US AI competition will continue, but ensuring that AI is used for good and benefits all humanity is the greatest common interest of China and the United States in this field. Should we "install guardrails" and "apply brakes" for technological development, or "engage in competition" and "pull down an iron curtain"? As major AI powers, China and the United States bear special responsibilities and should demonstrate responsibility through cooperation, jointly ensuring that AI technology brings not "the dusk of humanity" but "the dawn of a new era."