The White House has been arguing lately.

The trigger was a proposal that has not yet gone anywhere: a senior official suggested that AI companies should need government approval before releasing a new model. A director of the National Economic Council compared the idea on the spot to the FDA approval process for new drugs.

Allies of the AI companies immediately erupted: this would use bureaucracy to kill innovation.

I do not want to debate who is right in this internal fight. I want to talk about something else: when a government seriously considers regulating AI like medicine, there is a signal behind it that deserves more attention than regulation itself.

Comparing AI to medicine is an admission

What needs a “license to go on the market”?

Medicine. Because medicine can heal people, and it can kill them.

When a country reaches for the FDA model to frame AI, it may look like a discussion about regulatory procedures. In substance, it is an implicit declaration: AI is something with potentially toxic effects, and we are not sure it is safe for people.

It has taken longer than we expected for this comparison to arrive.

When ChatGPT first took off, the public response was mostly amazement. AI could write code, make pictures and draft your emails. It was a light, buoyant excitement: “Wow, this thing is amazing.” Politicians and regulators talked about the need to govern AI, but most of the time it felt like chasing a hot topic and saying a few things for show.

That has changed. The moment people in a White House meeting room seriously discuss an approval process before AI can be released means that some of them have begun to be genuinely afraid.

Not fear as a performance. A real belief that this thing might get out of control.

Every generation of new technology reaches this point

Nuclear power has been through it.

After it appeared, humanity used nuclear energy to generate electricity and also built weapons capable of destroying cities. Countries then signed the Treaty on the Non-Proliferation of Nuclear Weapons, established the International Atomic Energy Agency, and designed a long approval system: nuclear materials must be tracked, and nuclear facilities must be inspected.

Biotechnology has been through it too.

When CRISPR gene editing first appeared, scientists were both excited and alarmed. For the first time, people had a tool for “rewriting the code of life.” Today, the international community has clear boundaries around germline gene editing, and every new kind of biological product has to go through rigorous clinical trials.

The pattern is the same: when a technology shifts from “an exciting possibility” to “an unsettling reality,” people start using institutions to tame it.

AI is entering that phase. Not because it suddenly turned bad, but because it has become real enough, ordinary enough and influential enough to affect enough people. That is what makes us feel its weight.

Fear is where domestication begins

Some will say regulation slows AI progress.

That is true, of course. An FDA-style process would mean queuing for approval before every new model launch. Efficiency would fall, the pace of innovation would be disrupted and new opportunities for rent-seeking might emerge.

But there is another side: a technology constrained by nothing may ultimately be trusted by no one.

We trust planes because they have airworthiness certificates. We trust medicines because we have clinical trial data. We trust food because ingredients are listed. Those “annoying approval procedures” are precisely what allow the public to accept a new technology in everyday life.

If AI is really to become part of civilization’s infrastructure, sooner or later it will need a trust system that lets people dare to rely on it.

The argument in the White House has not been settled. The proposal may move forward, collapse or reappear in some compromise form.

But one thing has already happened: people have begun to take their fear of AI seriously, instead of just talking about it.

That is not a bad thing. The moment people become seriously afraid of a technology is often the moment they begin taking it seriously.

Sources: internal discussions on AI regulation at the White House, May 2026; pre-deployment evaluation agreements signed by the U.S. Department of Commerce’s NIST with Google DeepMind, Microsoft and others.