"The United States will win this race," U.S. President Donald Trump declared at the AI Summit on July 23, 2025, "because we will never allow any foreign country to defeat us."
Hot Topics
On the same day, the White House released the "American AI Action Plan," with "Winning the AI Race" in the title. A month later, China's State Council released its "Artificial Intelligence Plus" guidelines, proposing to promote AI in areas such as industry, consumption, people's livelihood, and governance, while emphasizing security capability building and global governance cooperation.
Both the US and China are accelerating their AI development, but calls for a slowdown are emerging from within the industry. In September 2026, Dario Amodei, CEO of Anthropic, the developer of Claude, published an article advocating for slowing down the pace of advancements in cutting-edge AI capabilities to allow security measures time to catch up. However, he also emphasized that a coordinated slowdown should preserve the leading edge of the US and its allies.
If the risk could affect everyone, why is it difficult for the party that values safety to slow down first?
Brian Patrick Green, director of technology ethics at the Markkula Center for Applied Ethics at Santa Clara University, said in an interview that competition makes developers worry, and caution means losing an advantage.
“Some people want to do things safely, but they feel that if they try to do things more safely, they will lose the race; those who are less safe will win,” Green said.
There are still disagreements about the risks, and which boundaries need to be set in advance?
In February 2026, the "International AI Security Report," led by computer scientist Yoshua Bengio and authored by more than 100 experts, described an out-of-control scenario as an AI system escaping human control, with the cost of regaining control being extremely high, or even impossible. The report concluded that, at its assessment point in time, the system did not possess the capabilities necessary to cause such an out-of-control situation; however, there was considerable disagreement among experts regarding the likelihood of this occurring in the future.
Green's concerns include both the potential for humans to use AI to commit harm and the potential problems that may arise if AI is given greater autonomy in the future. He cited examples such as criminal organizations, malicious individuals, and even nations attempting to use AI to hack into bank accounts or engage in other harmful activities. However, he emphasized, "Those bad outcomes are not necessarily going to happen."
Regarding longer-term risks, he suggested that if someone were to allow more capable AI to control biological laboratories in the future, it would be necessary to consider whether it could be used to create dangerous biological weapons.
“This is the more distant future, but we need to decide now what things we don’t want to happen,” Green said.
Will a safer product lose out to a faster competitor?
For competing countries and companies, security requirements also involve the pace of research and development and market launch.
In an interview, Li Houchen, a columnist specializing in current affairs and economic analysis, linked China's approach to promoting AI development to its industrial policies. He believes the government can drive development across multiple stages, from lithography machines and data centers to model development, procurement, and applications.
He said, "If they know someone is chasing them from behind, they'll feel the need to go faster to stay ahead."
In Green's view, overcoming this pressure requires participants to collectively accept safety baselines, share safety information, and coordinate development paces. He advocates for the establishment of international treaties that clearly define minimum safety standards for AI, including which tasks must be completed safely and which behaviors should be prohibited.
From communication to verification, how can we confirm that the other party is keeping their promise?
New communication arrangements have emerged between the US and China. According to a White House statement dated September 25, 2026, and an interview published on October 1 by Xie Feng, the Chinese ambassador to the US, both sides agreed to continue dialogue on AI-related risks and benefits and to establish communication channels for AI-related issues.
In Green's view, even if all parties agree to abide by the safety rules, there is still an enforcement problem to be solved: who will inspect them?
“Whenever a treaty is signed with someone, it needs to be ensured that it is followed, which requires a verification mechanism,” Green said. He envisions that future AI agreements may require international personnel to enter the relevant countries and companies to check whether the models have the agreed-upon security measures.
If AI causes harm, who should be held responsible?
In addition to reviewing safety commitments, Green also advocates holding the companies that developed the technology and the individuals involved accountable for any harm caused.
“If an AI system harms someone, then the creator of the AI should be held accountable,” he said. “You can’t hold the AI itself accountable, and there’s no indication that it has any moral responsibility.”
In his ethical judgment, the discussion should not stop at "the system has a problem," but should also delve into who made the choices regarding development, deployment, and use. He suggests considering holding specific decision-makers accountable to encourage greater emphasis on security within the company.
“I think one way to get people to take ethics and safety more seriously is to link responsibility to specific individuals,” Green said.
For Green, the safety issues in the AI race ultimately need to be addressed through executable arrangements: which boundaries cannot be crossed, who is responsible for oversight, and who is held accountable if harm occurs.
In the next episode, we will turn our attention to individuals and discuss how AI is changing the power relationship between governments and ordinary people.
