Hiroshima and Nagasaki disasters offer lesson for AI governance

TOKYO — Restricting artificial intelligence while still harnessing it for human prosperity has emerged as one of the 21st century’s greatest challenges. U.S. President Donald Trump and his Chinese counterpart, Xi Jinping, agreed at their summit on Sept. 24 to further discuss the issue.

The superpowers have different strategic calculations regarding AI. The U.S., which currently leads the technological race, is cautious about hindering it with regulations, while China hopes to catch up by taking the lead in setting rules that work to its advantage. This is why Xi emphasized the need for AI governance during his visit to the U.S.

Opinions are divided even among Western countries. One camp argues that the pace of AI development should be slowed, with priority placed on creating safety measures to prevent catastrophic outcomes. The other contends that development must accelerate to keep from losing ground to China, a view strongly favored by Trump.

But the West should strive to achieve both goals simultaneously; this is not impossible.

If priorities are set, the top one should be to impede AI from running out of control. Even if the West were to win its competition with China, that victory would be meaningless if AI-driven biological or chemical attacks, or the destruction of critical infrastructure, were to occur.

The history of nuclear weapons demonstrates how difficult it will be to control AI-related risks.

In June 1945, scientists involved in the development of the atomic bomb produced the Franck Report, which argued that an “unannounced attack against Japan” was inadvisable to avoid sacrificing public support for the U.S. throughout the world. Yet in August of that year, bombs were dropped on Hiroshima and Nagasaki.

After World War II, the U.S. and Soviet Union engaged in a nuclear arms race. Serious efforts at arms control only began after the 1962 Cuban Missile Crisis that had brought the world to the brink of nuclear war. The following year, the countries established a hotline, and the Nuclear Non-Proliferation Treaty entered into force in 1970.

It took 25 years after the atomic bombings until a global safety framework was established.

That much time is not left in the AI era. AI models can be replicated quickly, and once they spread, recalling them becomes difficult. Their owners are not limited to governments but also include companies, universities and individuals. If abused for military or terrorist purposes, the consequences could be catastrophic.

China is also concerned about uncontrolled AI, as indicated by one of its military officials at the Shangri-La Dialogue security summit in Singapore in late May. The militarization of emerging technologies, including AI, “is challenging traditional rules of engagement and ethics of war, increasing the risk of losing control,” Major General Meng Xiangqing said.

At the U.S.-China “Track 1.5” dialogue of government and industry officials held in Beijing in late August, AI safety was a central topic.

A framework to prevent AI-related disasters would benefit both the U.S. and China. The countries should establish mechanisms to ensure that major accidents and cyberattacks can be reported quickly, and share information on misuse by terrorist and criminal organizations.

Nevertheless, the West cannot afford to lose the AI race to China. Chinese AI systems have been widely criticized for refusing to answer questions about human rights issues or Taiwan, and for echoing Beijing’s official positions.

If such AI systems become deeply embedded in public administrations, schools and media around the world, the Chinese government’s narrative could gain greater acceptance around the world. China is also striving to create an international order for the technology by launching the World Artificial Intelligence Cooperation Organization.

Yet it is difficult to exclude Chinese AI entirely. Many AI models from the country are inexpensive and highly customizable, making them increasingly popular, particularly in emerging countries. If the U.S. blocks them from its market and those of its allies, Chinese models would only dominate the remaining nations.

As such, it would be wise for the U.S. and its Western allies to work together to develop AI systems that outperform Chinese alternatives and are easier to use.

The U.S. enjoys a significant lead in cutting-edge AI models and cloud infrastructure. Among its allies, Japan has manufacturing strengths that may prove advantageous in physical AI. Taiwan and South Korea are indispensable for semiconductor production, while Japan and Europe can play key roles in manufacturing equipment.

Maintaining U.S. leadership requires building AI models that its allies can trust.

In June, the U.S. Commerce Department temporarily restricted access to the latest model developed by Anthropic for national security reasons. If Washington is going to step in like this, it would help if it could inform the industry’s customers when access to AI models could be restricted and for how long.

With this kind of transparency, the U.S. could build the confidence it will need as it takes on China in the field. Cooperating with China to prevent AI-related disasters and preventing China from achieving AI hegemony are not mutually exclusive objectives.

In the case of nuclear weapons, the devastation of Hiroshima and Nagasaki, followed by the Cuban Missile Crisis, resulted in a globally shared understanding that unchecked escalation could lead to catastrophe.

“As for AI, such a sense of crisis is not yet widely shared, making it difficult [for countries] to agree on a comprehensive regulation,” said Nobumasa Akiyama, a professor at Hitotsubashi University and an expert on nuclear arms control. “For now, practical measures would be to create a system for accident reporting and a framework for crisis management.

“More important than the specific mechanisms, however, is confirming that countries in conflict can cooperate to protect their common interest of avoiding catastrophe.”

Major power rivalries hindered progress toward nuclear disarmament in the quarter century from the atomic bombings of Japan to the ratification of the NPT. Now similar rivalries are getting in the way of safeguarding humanity from the potential risks of AI.

Mechanisms to contain these risks must be established before they bring us to the edge of disaster.