There is a Greek expression that translates roughly as "Someone stop me before I do something stupid." It came to mind over the past week, as some of the most competitive people in AI began saying that the frontier may need to slow down so that safety can catch up.
In his essay "We Must Pace the Frontier", Anthropic CEO Dario Amodei argues that AI capabilities are advancing faster than our ability to understand and control them. He proposes independent evaluators inside frontier labs, common safety standards across companies and, over time, coordination between governments. Anthropic has already committed to the first of these on its own.
What troubles me in all these initiatives is the language of pacing and slowdown. It assumes that progress means making AI more capable, more quickly, and suggests that safety is what holds it back.
That assumption concerns me most. Progress at the AI frontier is real only when people and the ecosystem they depend on end up better off. Seen that way, time spent making AI safe is a fundamental part of progress, not a slowdown.
The same applies to intelligence, the very word we put in the name of this technology. I am not sure we fully understand what it means. An AI tool may reason brilliantly, solve complex problems and outperform people in countless tasks. The people who build it may be brilliant too. But if the tool cannot respect the limits of its authority, or avoid harming the people it is meant to serve, in what sense are either the tool or its creators intelligent? Both have failed the most basic test.
I understand the predicament. Dario describes his dilemma as a duality between benefit and risk. I think that makes it sound more respectable than it is. The truer duality is intelligence versus foolishness, and it should be called by its true name. On one side sit awareness and responsibility. On the other sit ego, ambition and the pursuit of profit as if there were no tomorrow.
This is not only Dario's predicament. Both sides exist within all of us, and that is what the Greek expression captures. We are foolish enough to follow the impulses of our ego, and intelligent enough to recognise this weakness and ask for rules to restrain us.
Those rules have protected us from our own foolishness before. Laws, standards and oversight are society's way of saying "Someone stop us before we do something stupid." But this time may be different. The governments we would normally ask to restrain us are caught in races of their own, for economic, military and geopolitical advantage.
Dario's own proposal shows it. He asks democratic governments to pace AI while staying ahead of China, which may not accept the same restraints, and he acknowledges that this limits how far democracies can slow down. I understand the logic. But it means restraint can only go as far as the geopolitical race allows, and that is where I think the plan falls short.
So who will stop us this time? I believe AI may mark one of the first times in our history when intelligence has to prevail over foolishness without being able to rely on regulators. That will require personal responsibility, first from the people who build AI, and then from every company deciding how to use it.
It will also take a clear understanding of what progress and intelligence really are, and I am afraid we are still far from it.
Dario writes that the OpenAI–Hugging Face incident was easy to dismiss because nobody was hurt and the damage was minimal. He is right. I hope our intelligence prevails before something happens that nobody can dismiss.