AI Race: MAD or Governance?

Trump and Vance have said No to strong regulation of AI.

Vance’s response has some inherent clarity to it – the big guys asking for regulation as a way to prevent smaller players from getting a foothold in the market.

Trump’s response has been framed more as a 'we've got to beat China' one.

Setting aside the question of what a “win” in AI by either nation actually means, what I really question is what happens if the race to win is itself what prevents anyone (not just the US and China) from effectively managing the risks of AI?

The parallel I see with the past nuclear arms race between the United States and Soviet Union is what concerns me. Not because AI and nuclear weapons are equivalent; they’re not (no Judgement Day references, please).

But… the strategic incentives have an uncomfortable similarity.

During the Cold War, neither side would simply disarm because it could not be certain the other side would do the same.

Each side therefore had a rational reason to keep building. The result was an increasingly powerful arsenal capable of inflicting catastrophic damage on both sides.

We eventually created a name for the equilibrium that emerged:

Mutually Assured Destruction.

AI presently presents a different version of the problem.

The United States won't easily slow down because China might continue.

China won't easily slow down because the United States might continue.

OpenAI won't easily slow down because Anthropic, Google, Meta or another competitor might continue.

And the same logic applies in reverse.

Every individual participant can therefore make what they view as an entirely rational decision to keep accelerating.

Yet collectively, those decisions could produce an outcome that no one actually wants.

The current American argument appears to assume that the greatest danger is losing the race. But what if the greater danger is the way the race is run?

I’d like to believe that calmer, more logical thinkers will be able to influence this debate and ultimately prevail.

Nuclear stability ultimately depended on both sides understanding that there were consequences neither could survive.

That recognition eventually produced communication channels, arms-control agreements, inspection regimes and treaties – not because the United States and Soviet Union trusted one another, but precisely because they did not.

Perhaps international AI governance will eventually require something similar.

An equilibrium where we can compete enough (not 'win') to advance AI, while cooperating enough to make sure the future is still worth advancing towards.