AI Made It Possible · Tech Culture & Power · Current Intelligence

The AI Race Has a Safety Problem: What Happens When No One Wants to Slow Down First?

The United States and China are competing to build more powerful AI while discussing ways to stop that competition from producing a crisis. The real question may no longer be “race or regulate?” It may be “how do you do both at once?”

Gary Whittaker · Founder and Operator, JackRighteous.com · September 23, 2026

For years, the AI debate has often been framed as a choice between moving fast and being safe. I increasingly think that is the wrong frame.

The United States and China are both treating advanced AI as strategically important. At the same time, officials and security experts from both countries are discussing communication channels, human-control rules and safeguards designed to reduce the chance that AI-related incidents escalate into something much worse.

Trump made the American position unusually clear

On September 22, President Donald Trump told the United Nations that the United States rejects attempts to create what he described as a global system of control over artificial intelligence. In the same speech, he said U.S. government documents would begin referring to AI as “Super Intelligence.”

The wording was dramatic, but the larger policy direction is familiar: the United States increasingly treats AI leadership as an economic, technological and national-security priority.

Read the White House release from Trump’s September 22 UN address →

That makes “just slow down” harder than it sounds. If one country believes advanced AI could create a major strategic advantage, slowing development only works if it trusts competitors to do the same.

That trust does not currently exist.

China is racing too — but it is not proposing the same governance model

This is where I would not repeat the easy headline that both governments have simply chosen “no brakes.”

China is developing AI aggressively and treating it as central to national industrial and technological strategy. But its public governance language is different from Washington’s. Chinese officials have promoted international coordination, sovereignty, development and safety through mechanisms including the World Artificial Intelligence Cooperation Organization.

That does not mean China is slowing down. It means the competition is happening on two levels at once:

The race is not only about who builds the best AI. It is increasingly about who gets to define what responsible AI looks like.

The strange part: rivals are also discussing safety together

U.S. and Chinese officials have been discussing continued AI-safety dialogue, including the possibility of direct notification mechanisms for serious AI incidents. Separate U.S.–China expert discussions have proposed safeguards inspired by nuclear risk management: clearer red lines, meaningful human control over critical military systems and a hotline for AI-related crises.

Reuters: U.S.–China security experts propose nuclear-style safeguards for AI risks →

The nuclear analogy is becoming less metaphorical

I have used the nuclear analogy before when discussing AI — not because AI is literally a nuclear weapon, but because both technologies can create enormous benefits and enormous consequences depending on how they are controlled.

Now policymakers and security researchers are beginning to reach for some of the same mechanisms that emerged during the nuclear era: hotlines, red lines, explicit human control and crisis communication.

The United Nations made the comparison directly this week. In opening remarks for the General Assembly, the president of the assembly warned that the world cannot allow an arms race for control of AI to carry humanity “over the precipice,” while also saying innovation should continue.

Read the September 22 UN General Assembly remarks →

That sentence captures the contradiction almost perfectly:

Keep innovating. Build guardrails. Do not lose the race. Do not let the race become catastrophic.

But AI is harder to control than a nuclear arsenal

Nuclear weapons depend on highly specialized materials, infrastructure and state-level capabilities.

AI is different.

Frontier systems are expensive to train, but capability spreads through models, APIs, cloud access, open weights, research, software, distillation, chips and consumer products.

That means AI cannot be governed simply by counting systems or locking a physical stockpile in one location.

It is already embedded in businesses, health systems, militaries, software and everyday tools.

And the risk evidence is more complicated than the loudest headlines

I do not think the useful position is “AI is going to kill us” or “AI risk is fake.”

The evidence supports a more disciplined position.

Current systems have not demonstrated the full capability required for the most extreme loss-of-control scenarios. At the same time, autonomy, planning, tool use and evaluation-avoidance are improving enough that serious researchers are studying the risks now rather than waiting until the systems are dramatically more capable.

That is exactly why I keep returning to one rule throughout this series:

Call demonstrated capability what it is. Call forecasts forecasts. Do not confuse uncertainty with proof in either direction.

For the deeper evidence standard behind that position, read If AI Is Going to Kill Us, Show Me the Evidence and Superintelligence: How Would We Actually Know?.

The real trap is incentives

Imagine both countries believe future AI could create serious risks.

That still does not mean either country has an incentive to stop first.

Each side has to ask:

That is the heart of the problem.

Safety may be desirable collectively while acceleration remains rational individually.

Once AI becomes tied to military capability, scientific leadership, economic productivity and global standards, the incentive problem gets harder.

The public is not necessarily asking the same question governments are

A Reuters/Ipsos poll released September 22 found that 55% of U.S. adults supported slowing AI development. In the same poll, 73% prioritized safe and responsible development over global competitiveness when asked to choose between the two. The online survey included 1,277 U.S. adults and had a margin of error of roughly three percentage points.

Reuters/Ipsos: U.S. public concern about AI safety →

That does not tell governments what policy they should adopt.

But it reveals another tension:

This is why the human-control question matters more than the branding

The day before writing this, I examined Trump’s proposal to rename AI “Super Intelligence.” My concern was that language can make a technology easier to anthropomorphize at exactly the moment when we need clearer accountability.

This geopolitical story strengthens that point.

Who controls the systems?

Who grants access?

Who connects them to weapons, networks or infrastructure?

Who decides when a system can act autonomously?

Who is responsible when incentives reward speed over caution?

Those questions do not disappear because models become more capable.

They become more important.

That is also why this article belongs beside AI Made It Possible, Part II: The Fight Was Never About the Machine. The machine matters. But the humans, institutions and incentives around the machine still matter too.

AI can still be an empowerment story

I remain deeply optimistic about what AI can make possible for ordinary people and independent creators.

People who once lacked access to production, programming, publishing, music, visual tools or large teams can now attempt projects that would have been unrealistic only a few years ago.

That matters.

But optimism about a technology does not require pretending there are no risks attached to it.

AI can expand human capability.

AI can create meaningful new risks.

Governments can need safeguards.

Governments can simultaneously have incentives to accelerate.

Companies can build extraordinary technology.

Companies can still require oversight.

Those statements can all be true at the same time.

The race is already here

The United States and China are competing over AI capability, infrastructure, chips, models and standards.

They are also discussing ways to keep certain AI failures from escalating into larger crises.

That tells us something important.

The world may not be choosing between competition and safety.

It may be trying to build safety inside competition.

Whether that works is one of the most important technology-policy questions of the next few years.

Gary Whittaker
Founder and Operator, JackRighteous.com

Create What You Love | Love What You Create.