The risks are substantial, but we can resolve them by thinking about safety from the first day.
For nearly a decade, Google's Sundar Pichai has stood at the intersection of technological power and moral responsibility, urging the world's governments to treat artificial intelligence not as a weapon to be wielded but as a commons to be governed. Speaking first at Davos in 2018 and repeating the call ever since, he has invoked the Paris Climate Agreement as a model — a reminder that humanity has, at least once, chosen coordination over competition when the stakes were civilizational. Seven years on, as AI grows more capable and more embedded in the world's critical systems, the window for that choice may be quietly closing.
- Pichai's warning, first issued when AI risks were still largely theoretical, now lands in a world where those risks are becoming operational realities embedded in infrastructure and governance.
- Despite the urgency, the international community has produced no binding framework for AI governance, and the competitive logic Pichai cautioned against — nations racing for military and strategic advantage — has largely set the pace.
- Google itself continues advancing its AI capabilities with the forthcoming Gemini 3.0, creating a tension between the company's advocacy for safety-first development and its own drive to remain at the frontier.
- The Paris Climate Agreement analogy cuts both ways: it represents what collective action can achieve, but also how slowly and imperfectly nations move when economic and strategic interests are at stake.
- The critical question has shifted — no longer whether AI should be demilitarized, but whether the moment for achieving that agreement has already begun to pass.
Sundar Pichai has spent nearly a decade issuing a warning that has only grown more urgent with time. When he took the stage at the World Economic Forum in Davos in 2018, he called artificial intelligence "probably the most important project humanity has ever worked on" — placing it above electricity, above fire. The statement was not conference theater. It was a measured alarm, because alongside that assessment came a stark recognition: if nations treat AI as a domain for military competition rather than collective stewardship, the risks will compound in ways that become impossible to reverse.
What Pichai proposed was international demilitarization — a binding framework modeled on the Paris Climate Agreement, in which nearly every nation committed to coordinated action on climate change. AI, he argued, needed the same architecture: shared standards, transparency requirements, and a safety-first commitment from the earliest stages of development. The logic was clear. A country that unilaterally accepts safety constraints while others do not is, in theory, at a disadvantage. The agreement had to be mutual to be meaningful.
Seven years later, the world has not moved in that direction. Few governments have established common regulatory frameworks. Fewer still have approached anything resembling a demilitarization accord. The competitive logic Pichai warned against has largely prevailed, even as Google itself continues advancing its own capabilities — the company is now preparing to launch Gemini 3.0, the next generation of its large language models.
Pichai has maintained throughout that transparency and proactive safety thinking are the responsible path forward. But the international response has remained sparse, and the technology has grown far more capable and far more integrated into critical systems than it was when the warning was first issued. The question that once seemed philosophical — should nations demilitarize AI? — has become urgently practical: do they still have the time, and the will, to try?
Sundar Pichai has spent the better part of a decade warning the world about artificial intelligence—not as a technologist marveling at what's possible, but as someone who believes the stakes are existential. In August, he marked ten years as CEO of Google, a tenure defined by navigating the company through regulatory scrutiny, competitive pressure, and the accelerating arrival of AI systems that are reshaping how information moves through human civilization. But the challenge that has held his attention most persistently is not a quarterly earnings target or a market share battle. It is the question of whether nations will treat artificial intelligence the way they treat nuclear weapons, or whether they will let it become one.
Back in 2018, speaking at the World Economic Forum in Davos, Pichai made a statement that has only grown more pointed with time. He called artificial intelligence "probably the most important project humanity has ever worked on." He placed it above electricity, above fire—the foundational technologies that remade human civilization. That framing was not hyperbole meant for a conference audience. It was a warning dressed in measured language. Because alongside that assessment came a stark recognition: the risks embedded in AI's development are substantial, and they will only compound if nations treat the technology as a domain for competitive military advantage rather than collective stewardship.
What Pichai proposed was a form of international demilitarization. He argued that countries needed to work toward a common objective: preventing artificial intelligence from becoming a weapon in the traditional sense, or worse, a tool for asymmetric advantage in conflicts yet to come. To illustrate what he meant, he pointed to the Paris Climate Agreement—the 2015 treaty in which nearly every nation on Earth committed to addressing climate change through coordinated action and shared standards. AI, he suggested, needed something similar: a binding international framework, transparency requirements, and a commitment to safety-first development from the earliest stages.
The logic was straightforward. A technology with the potential to reshape economies, influence information flows, and amplify human capability at scale cannot be left to the whims of individual nations racing to build the most powerful version first. The incentives are misaligned with the risks. A country that unilaterally commits to safety guardrails while others do not is, in theory, at a disadvantage. That's why the agreement had to be international, binding, and mutual. That's why Pichai kept returning to it.
Yet seven years have passed since that warning. Google itself has continued advancing its own AI capabilities—the company is now preparing to launch Gemini 3.0, the next evolution of its family of large language models. The search engine that made Google's fortune has had to contend with new competitors and new regulatory pressures. The company has had to manage the demands of innovation in operating systems, cloud infrastructure, and the broader ecosystem of AI-dependent services. Through it all, Pichai has maintained that transparency and forward-thinking about safety are the paths to managing the risks.
But the international response has been sparse. Few governments have established common regulatory frameworks for AI. Fewer still have moved toward anything resembling the kind of demilitarization accord Pichai described. Instead, the world has watched as different nations pursue their own AI strategies, with varying degrees of caution and varying levels of investment in safety research. The competitive logic that Pichai warned against—the logic that treats AI as a domain for national advantage—has largely prevailed.
What remains unclear is whether the warning will be heeded before the technology reaches a threshold where international agreement becomes much harder to achieve. Pichai's call for cooperation was made when AI was still in its early stages, when the risks were theoretical enough to discuss calmly at Davos. Now, as AI systems become more capable and more integrated into critical infrastructure, the window for establishing shared rules may be narrowing. The question is no longer whether nations should demilitarize AI. It is whether they still can.
Notable Quotes
The IA is probably the most important project humanity has ever worked on. I consider it something more profound than electricity or fire.— Sundar Pichai, World Economic Forum 2018
Countries need to demilitarize AI. That is a common objective that countries should work toward.— Sundar Pichai, World Economic Forum 2018
The Hearth Conversation Another angle on the story
When Pichai compares AI to electricity and fire, is he saying the technology itself is neutral—that it's just a tool?
No. He's saying the scale of impact is so vast that neutrality becomes impossible. Electricity and fire transformed human society. AI will do the same, but faster, and in ways we can't fully predict. That's why he keeps emphasizing the need for international agreement before the competitive logic takes over.
But Google is building Gemini 3.0 while making this argument. Doesn't that create a contradiction?
It does, and Pichai seems aware of it. He's arguing that Google should lead responsibly while also pushing governments to create the conditions where responsible development is the norm, not the exception. The contradiction only disappears if everyone agrees to the same rules.
Why hasn't his message landed? Seven years is a long time.
Because demilitarization requires countries to trust each other and to accept constraints on their own capabilities. That's a hard sell when the technology is advancing so rapidly and the competitive advantage is so clear. It's easier to build than to agree not to build.
Is there still time to establish the kind of international framework he's describing?
That's the real question. The longer nations wait, the more entrenched their AI strategies become, and the harder it is to coordinate. Pichai's warning was loudest when AI was still emerging. Now that it's here, the political will to constrain it may have already passed.
What would demilitarization actually look like in practice?
Shared standards for safety testing. Transparency requirements about how AI systems are trained and deployed. International inspections, maybe. Agreements not to weaponize certain capabilities. It's the same logic as nuclear non-proliferation, but for a technology that's much harder to control because it's dual-use—it has civilian and military applications that are almost impossible to separate.