What if the most important invention in human history arrives before we agree on who inspects it?
That is the tension at the center of the current AGI debate.
Demis Hassabis, the co-founder and chief executive of Google DeepMind, has argued that artificial general intelligence may be only a few years away. AGI means more than a faster chatbot or a smarter search engine. It suggests a system with broad reasoning ability — something capable of learning, planning and solving problems across many domains in a way that begins to rival the human brain.
That milestone has not been reached. No laboratory has publicly demonstrated a universally accepted version of AGI. Researchers still disagree about how to define it, how to test it and what evidence would prove it had arrived.
So the timeline matters. Hassabis’ “few years” estimate should be treated as a prediction, not a certainty.
But predictions can still change policy. If the people closest to frontier AI believe the threshold may be near, society has to ask what kind of guardrails should exist before the next generation of systems is released.
Hassabis’ proposed referee
Hassabis’ proposal is a standards body modeled partly on FINRA, the private-industry watchdog that polices Wall Street. In the AI version, technical experts would evaluate powerful frontier models before public release. They would test for dangerous capabilities, review deployment risks and help coordinate a slowdown if a model appeared too risky to release.
The attraction of that idea is speed. Government regulation often moves slowly. Frontier AI does not. Models can gain new abilities in months. Labs compete intensely. Capital moves quickly. Public understanding lags behind the release cycle.
A specialized standards body could move faster than a traditional regulator. It could bring together technical experts, private labs and public agencies. It could create shared tests, common reporting standards and a neutral checkpoint between invention and mass deployment.
That is the promise.
The weakness is also obvious. A self-regulating body can become too close to the industry it is supposed to oversee. If the same companies building frontier AI help fund or shape the referee, the public will want to know how independent that referee really is.
The FINRA comparison is useful because it shows both sides of the model. Industry expertise matters. But industry influence is always a risk.
The second race
The AGI debate is often framed as a race between companies or nations. Who gets there first? Which lab has the best model? Which country controls the most advanced systems?
But Hassabis’ proposal points to a second race: the race to build institutions capable of understanding and governing powerful AI before it becomes deeply embedded in everyday life.
That race may be just as important.
If AGI arrives later than expected, strong evaluation systems could still improve safety for today’s frontier models. If AGI arrives sooner than expected, the absence of those systems could become a historic failure.
The question is not only whether machines will become more capable.
It is whether human institutions can become capable fast enough.
Editorial note: AGI remains a hypothetical and contested milestone. Statements about its arrival within a few years should be attributed to Hassabis and presented as predictions rather than established timelines.
