If AI Is So Dangerous, Why Do They Keep Developing It? Actions Reveal What Words Do Not

“If AI continues to advance at this pace, it could pose a serious threat to humanity.”

Recently, AI companies themselves have repeatedly issued warnings like this.

The companies at the forefront of AI development, such as OpenAI and Anthropic, are among the most vocal about safety and the need for government regulation and international rules.

But I do not take this argument entirely at face value.

The reason is simple.

If they believe it is that dangerous, why don’t they stop developing it themselves?

If They Are Truly Afraid, They Could Stop

“This technology is dangerous.”

“It could cause irreversible harm to humanity.”

If they are willing to go that far, the most reliable safety measure would be to stop development at their own companies.

Yet no one leaves the race.

They build new models, construct enormous data centers, secure semiconductors, and compete for the best researchers. At the same time, they insist that AI is dangerous and international regulation is necessary.

In other words, what they are asking for is not:

“It is dangerous, so I will stop.”

It is:

“It is dangerous, so all of us should stop together.”

I suspect there are considerations here that cannot be explained by ethics alone.

Stopping Alone Means Losing

Even if OpenAI announced, “For the sake of humanity, we will suspend development for two years,” there would be no guarantee that Google, Anthropic, and Chinese companies would stop as well.

By the time OpenAI returned two years later, its competitors might be far ahead.

That is why no company can leave the race on its own.

Government regulation or international standards, however, could slow development across the entire industry without putting one company alone at a disadvantage.

And today’s AI companies have another reason to want the pace of development to slow.

Money.

AI Companies Are Burning Money at a Ferocious Rate

Developing AI today costs an extraordinary amount of money.

Semiconductors, data centers, electricity, networks, and researchers: as models grow larger, the infrastructure they require expands as well.

OpenAI has reportedly been projected to consume approximately $278 billion in cash between 2026 and 2030.

AI is both a glamorous industry of the future and an industry that consumes capital at a ferocious rate.

That leads me to wonder:

Perhaps they would like to calm the race down for a while.

Yesterday’s cutting-edge model becomes obsolete six months later, demanding even more GPUs and even larger data centers. No matter how much money they invest, the race never ends.

At some point, they may want to slow the competition, sell their existing models to businesses, expand adoption of their services, and recover the money already invested.

But if one company alone takes a break, it loses.

And so the message becomes:

“For the sake of AI safety, let us all slow down a little.”

International rules justified by safety concerns are also extremely convenient if they create time to recover investments.

When I hear AI companies discuss safety, I am not willing to ignore this economic motive.

The Other Player Is China

There is another major presence in this story.

China.

If strict safety requirements become international standards while American companies still lead in advanced AI, companies trying to catch up later will be forced to bear the same costs.

Build a large safety division. Evaluate models. Report to the government. Submit to reviews when using computing resources above a certain threshold.

Large corporations may be able to comply with rules like these, but they impose a heavy burden on new entrants and companies that are behind.

In other words, safety regulations may be rules that protect humanity, but they may also become powerful barriers to entry that benefit the companies already in front.

At the national level, they could also serve as a tool for slowing China’s pursuit.

The ethical language of “regulating dangerous AI” aligns remarkably well with the national interest of “preserving America’s technological advantage.”

But Will China Really Stop?

The problem is whether China would actually follow those rules.

AI is directly connected to the economy, national security, the military, cyber capabilities, and science and technology.

If China continues developing while the United States spends time checking compliance with safety standards and regulations, the gap could close rapidly. The reverse is also true.

Without a guarantee that the other side will stop, neither side can afford to stop alone.

Ultimately, this becomes less a question of ethics and more a question of game theory.

Even if slowing down would make everyone safer, slowing down alone becomes the most dangerous option when there is no guarantee that the other side will do the same.

It Is Not as Simple as Nuclear Weapons

Moreover, AI is different from nuclear weapons.

Nuclear weapons involve physical objects and facilities—nuclear material, enrichment plants, missiles, and nuclear tests—which can be monitored, even if imperfectly.

AI is more ambiguous.

Researchers, algorithms, computing resources, model weights, and data centers: even the boundary between civilian and military use is unclear.

Nor is there an international order in which secretly developing AI would immediately expose a country to military attack by other nations.

Even if an international safety agreement were created, suspicion that one country was secretly racing ahead would cause the others to press the accelerator too.

And the competition would begin again.

Actions Reveal True Intentions

I cannot see inside the minds of AI company executives.

So I look not at what they say, but at what they do.

They say AI is dangerous to humanity while continuing to build new models.

They say development must slow down while refusing to leave the race.

They say international rules are necessary while declining to stop their own development first.

To me, those actions speak more eloquently than their words.

That is why I am unwilling to judge the situation solely by hearing that “we must make AI safe for humanity.”

If you want to know what a company truly believes, watch its behavior instead of reading its statements.

True intentions are revealed not in words, but in attitudes and actions.

And so far, not one of the companies warning most loudly that AI is dangerous has withdrawn from the AI development race.