The AI Non-Proliferation Treaty: Will Safety Become a Tool of Power?

The race to build the world's most powerful artificial intelligence systems was once Silicon Valley's greatest obsession. Companies competed to create larger models, acquire more computing power and move faster than their rivals. Speed was celebrated as progress.

Now, something unusual is happening. Some of the same companies that accelerated this race are warning the world that it may be moving too quickly.

The leaders of three of the most influential AI companies, Dario Amodei, Sam Altman and Elon Musk, have expressed support for slowing the development of frontier AI systems. Amodei's call to "pace the frontier" argues that AI capabilities are advancing faster than society's ability to manage their consequences.

The concern is not imaginary. AI systems are moving beyond simple chatbots. Increasingly autonomous AI agents can write code, interact with digital systems and perform complex tasks with limited human supervision. Failures could trigger cyberattacks, misinformation, economic disruption and even national security risks.

But beneath these legitimate concerns lies a more uncomfortable question: when those who won the AI race ask the world to slow down, are they protecting humanity or protecting their technological advantage?

The debate over AI safety is therefore not only about technology. It is about power.

History offers a warning.

When nuclear weapons emerged, the world faced a similar dilemma. The United States and the Soviet Union eventually accepted arms-control agreements. These reduced the danger of uncontrolled proliferation but also created a hierarchy in which a handful of countries retained nuclear capabilities while others accepted restrictions.

The contradiction was clear: the technology was considered too dangerous for many countries, yet acceptable in the hands of those who already possessed it.

The AI era risks creating a similar imbalance.

Today's leading AI companies have amassed enormous advantages through advanced chips, massive computing infrastructure, investment and elite talent. After racing to establish dominance, they are now advocating stronger global rules, independent safety evaluations and coordinated limits on AI development.

Their concerns deserve attention. Advanced AI systems may eventually make autonomous decisions with consequences governments are not prepared to manage. But regulation designed primarily by today's technological leaders carries another danger: it could become a gatekeeping mechanism.

History shows that powerful countries and corporations often champion openness while catching up and restrictions once they become dominant. The same pattern has appeared in nuclear technology, advanced semiconductors and other strategic industries.

The battle over AI is already intertwined with semiconductor geopolitics. US restrictions on exporting advanced chips to China reflect genuine security concerns, but they also shape who can build the next generation of AI.

A global AI governance framework is necessary. The question is who will write the rules, and whose interests will they protect?

If frontier AI remains concentrated in a handful of corporations and countries, much of the developing world risks entering the AI age as consumers rather than creators. Instead of narrowing inequalities, AI could deepen a new form of technological dependence.

This does not mean rejecting regulation. The world needs transparent safety standards, independent evaluations, accountability and international cooperation. Powerful technologies cannot be governed by market competition alone.

But safety cannot become a new language for technological protectionism.

The AI race also differs from the nuclear race in one crucial way. Nuclear weapons were controlled by governments. AI power is increasingly concentrated in private corporations whose decisions affect billions but are not directly accountable to democratic institutions.

The answer is neither an uncontrolled AI race nor a closed technological club where today's leaders decide tomorrow's limits. The challenge is building an AI order that combines innovation with fairness, safety with openness and competition with accountability.

Every transformative technology creates two races: one to build it and another to control who gets to use it.

The AI age will be no different.

The most important question is not only how powerful artificial intelligence should become or how quickly it should develop.

It is this: Will AI become a technology that empowers humanity, or a strategic asset controlled by those who arrived first?