Will AI’s new ‘mutual destruction’ be assured?

The past week was crucial for Japan’s government and business communities. On Sept. 22, Japanese Prime Minister Sanae Takaichi held talks with U.S. President Donald Trump in New York, followed two days later by a U.S.-China summit in Washington between Trump and Chinese leader Xi Jinping.

Ahead of the U.S.-China summit, there were concerns in Tokyo that Japan and the United States might have somewhat different priorities. Japan sought to confirm a shared understanding not only on the East and South China seas but also on peace and stability in the Taiwan Strait — specifically, a position that would not tolerate attempts by China to alter the status quo.

The U.S. side, meanwhile, appeared particularly focused on reaching at least a minimal understanding with China on the risks posed by artificial intelligence and their implications for military security.

Washington also had an interest in such issues as critical-mineral supply chains, including rare earths, as well as tariffs and trade. In Japan, however, there were concerns that the Taiwan issue and U.S. arms sales to Taiwan might receive lower priority.

As it turned out, the Japanese and U.S. leaders had a frank exchange on China, including economic security, semiconductors and artificial intelligence, and the meeting reaffirmed the strength of the U.S.-Japan alliance.

But my focus today lies elsewhere.

At the U.S.-China summit, neither side was likely to make major concessions on trade, rare earths or Taiwan. Rather — and this may not be widely understood in Japan — artificial intelligence could become the most consequential issue between the United States and China.

Moreover, I believe the summit could prove historically significant for reasons that go beyond restricting AI competition or deciding which advanced semiconductors can be exported.

AI is likely to become a pillar of strategic deterrence in the future, much as nuclear weapons have been. Although AI-related technologies remain in their early stages, I fear that within the next 10 to 20 years, advances far beyond what we can now imagine could turn the competition over AI into a contest for strategic advantage.

From the 1950s onward, the United States and the Soviet Union regarded nuclear weapons as the ultimate strategic weapon and pursued increasingly sophisticated arsenals. Eventually, the two nuclear superpowers developed a tacit understanding based on mutual deterrence, embodied in the concept of “mutually assured destruction,” or MAD.
I fear that AI could acquire comparable strategic value by the 2030s.

In such a world, the United States and China could eventually possess AI capabilities capable of overcoming or neutralizing the other’s strategic systems. This is why some AI experts have argued for restrictions on AI development to prevent such a scenario. But there is a historical parallel: Similar arguments were made about nuclear weapons development during the Cold War.

Neither the Trump administration nor Xi’s government appeared likely to agree to meaningful restrictions on AI capabilities at this stage. Both countries were racing to develop the technology, each hoping to secure a strategic advantage.

If that competition continues for another decade or two, the United States and China could acquire capabilities that would allow them to severely disrupt the other’s digital infrastructure and, potentially, its physical society. If so, discussions could emerge about an AI version of MAD: a system of mutual deterrence based on the recognition that neither side could safely attack the other without risking catastrophic retaliation.

Such a system could, paradoxically, produce a period of relative stability. Once both countries understood that continued AI development had created the potential for mutually assured destruction, the fear of catastrophic consequences could restrain their behavior.

But I would not assume that such a development would necessarily make the international system more stable. By then, China’s nuclear arsenal could exceed 1,500 warheads, potentially making it a third nuclear superpower alongside Russia and the United States. That would fundamentally complicate the existing U.S.-Russia model of nuclear deterrence.

The traditional logic of MAD rests largely on a bilateral relationship: If one nuclear power attacks the other, it can expect devastating retaliation. But a world with three nuclear superpowers would be more complicated.

If the United States, China and Russia each possess arsenals numbering in the thousands, the distribution of nuclear capabilities will not necessarily be equal. A tripolar nuclear balance could create strategic calculations that differ significantly from those of the U.S.-Soviet and later U.S.-Russia relationship.

In other words, even if a new form of MAD emerges in the AI domain, the traditional model of nuclear MAD could become less reliable in a three-power system. And the same problem could eventually emerge in AI. Rather than a bilateral U.S.-China competition, the world could face a more complicated tripolar AI balance involving the United States, China and Russia.

Consequently, even if discussions about AI at this U.S.-China summit remained preliminary, they could become a catalyst for developments that significantly shape the strategic balance among the United States, China and Russia over the next 10 to 20 years.

That is why the AI discussions at this summit deserve close attention. They are not simply about technology, trade or semiconductor exports. They could offer an early indication of how the great-power balance will evolve in the coming decades.

The question for Asian countries, including Japan, is whether they fully understand the strategic implications. Taiwan may not be the only issue that matters in the U.S.-China relationship.