Demis Hassabis has spent much of his career working toward artificial general intelligence. He now leads Google DeepMind, one of the organizations closest to the frontier, and believes AGI may be only a few years away.

That is why this line from his new essay stood out:

“Nobody in the world knows for sure what is going to happen from here, and even the experts disagree.”

We are already seeing early signs of the challenges posed by today’s frontier models, particularly in cybersecurity, increasingly agentic behavior, and our limited ability to understand why advanced models behave the way they do.

At the same time, commercial and geopolitical competition is pushing capabilities forward faster than our ability to fully evaluate, interpret, secure, and govern them.

Hassabis describes this period as a precious window before AGI arrives: a window to build the technical and institutional foundations we may soon depend on.

His proposal for a Frontier AI Standards Body is one attempt to do that. It would define capability thresholds, test the most advanced models before release, update evaluations as the frontier moves, and strengthen coordination among AI labs, government agencies, national laboratories, and independent auditors.

The exact framework will need debate. But the broader response to the essay was also notable.

Sam Altman, Satya Nadella, Sundar Pichai, Elon Musk, and others publicly described the proposal as thoughtful or important.

That does not mean they agree on every detail, or even on exactly when AGI may arrive. But it suggests that, among many of the people closest to the frontier, the need for stronger institutions is becoming a serious and immediate discussion.

The question is whether we use this window to build those institutions before AI capabilities move even further ahead of our understanding.