Before We Fear AI, Fear the Threat We Already Built

Turn on the news today, and AI is at the top of the news cycle. Artificial intelligence is having its moment, and it is difficult to find a major issue receiving more attention. Around the world, the conversation has a familiar refrain: AI could get out of control. It could threaten humanity. Slow it down, regulate it, cooperate globally and put guardrails in place before it is too late.

Everyone is talking about AI. The companies building it and the companies making the chips that power it are talking about its risks. Scientists, politicians, business leaders and governments are debating how fast it should advance, how it should be regulated and whether sufficiently powerful artificial intelligence could someday threaten humanity itself.

Those concerns deserve serious attention. But much of the proposed solution seems to depend upon one enormous assumption: that somehow the nations (and major AI corporations) of the world will come together, agree upon the danger and establish meaningful global rules to control it.

Are we kidding ourselves?

Look at our record. We already face global threats requiring exactly that kind of international cooperation, and we haven't solved them. Did I miss something, or did nuclear annihilation and environmental threats somehow go away while we were talking about AI?

Consider nuclear weapons. There is nothing theoretical about their destructive potential. We built them more than 80 years ago, demonstrated what they can do, and today nine nations possess about 12,200 nuclear warheads. About 4,000 are deployed with operational forces, with 2,100 to 2,200 kept on high operational alert.

Doesn't it seem obvious that nuclear weapons simply do not belong in a civilized world? Full stop.

These are weapons with a known capacity for destruction on a scale unlike anything else humanity possesses. Their danger isn't theoretical, nebulous or dependent upon predictions about what they might someday become. We know what they are capable of doing.

Yet humanity cannot agree to eliminate them.

After generations of diplomacy, treaties and negotiations, nine nations still possess nuclear arsenals. On February 5, 2026, New START—the last treaty legally limiting the strategic nuclear arsenals of the United States and Russia—expired without a successor treaty.

That brings us directly back to AI.

If the world cannot reach lasting agreement about the obvious danger of possessing weapons with the known capacity to devastate civilization through a single catastrophic decision, how can we realistically expect nearly 200 sovereign nations to reach enforceable agreement about something as nebulous and difficult to contain as software?

Climate change provides another example. Nearly every nation came together around the Paris Agreement, certainly a diplomatic accomplishment. But countries' emissions commitments are nationally determined, and the agreement's compliance process is explicitly non-punitive. Nations agreed that a global problem deserved a global response, but what emerged was an agreement largely dependent upon countries meeting their own commitments, without an international authority capable of enforcing their emissions targets.

Agreement is not enforcement. A conference is not compliance. Acknowledging a common danger is very different from accepting the economic, political or military sacrifices necessary to eliminate it.

AI makes that challenge even greater. AI isn't sitting inside a missile silo where an inspector can count it. It is software. It crosses borders electronically, can be copied or concealed, advances rapidly and promises enormous economic and military advantages. If one nation believes a competitor is developing substantially more powerful AI, will it voluntarily restrain its own development—or accelerate it?

There is another reason AI and nuclear weapons should be part of the same conversation.

AI doesn't have to push the nuclear button.

Humans can remain completely in control of that decision. AI merely has to influence the information upon which the decision is based.

Military and intelligence organizations must process enormous quantities of satellite imagery, radar information, communications and intelligence reports. AI could dramatically improve that process. But imagine a nuclear crisis in which leaders have only minutes to determine whether an apparent attack is real. An AI-supported system could misinterpret information, be deliberately deceived, make an innocent error or present an incorrect assessment with such apparent confidence that humans trust it.

The computer would not have launched anything. Human beings would still make the decision. But they could make the most consequential decision in history based upon information produced, filtered or interpreted by artificial intelligence.

Perhaps, then, we are asking the wrong question.

Instead of asking only whether AI might someday acquire the power to destroy humanity, perhaps we should also ask whether AI could make humanity more likely to misuse the destructive powers we already possess.

None of this argues against AI safeguards. Establish reasonable standards. Test advanced systems. Protect critical infrastructure and maintain meaningful human control over military decisions. International cooperation should certainly be pursued.

But we should be realistic about what it can accomplish. Nearly 200 sovereign nations will not suddenly cease having competing economic interests, military objectives and national ambitions simply because everyone agrees AI could be dangerous.

If the possibility of technological extinction has finally captured the world's attention, perhaps some of that urgency should also be directed toward the technology already capable of producing it.

The greatest near-term existential danger from AI may not be that it invents an entirely new way to destroy humanity. It may be that we connect an extraordinarily powerful new intelligence technology to the extraordinarily destructive technologies we have already created.

We are worried that AI might someday acquire the power to destroy civilization.

The uncomfortable truth is that humanity already has that power.