What if the biggest problem with AI is not just how capable it may become, but how quickly those capabilities are advancing?
In his essay We Must Pace the Frontier, Anthropic CEO Dario Amodei argues that frontier AI development should be deliberately paced. Not stopped, but slowed enough for safety research, independent evaluation, and public decision-making to keep up.
It is a remarkable proposal, especially from someone whose company is itself competing at the frontier. But it raises difficult questions.
Who decides what pace is still responsible? Who determines when a model is safe enough? How independent can evaluators really be if they depend on deep access to private companies? And can voluntary restraint work when commercial and geopolitical competition rewards those who move fastest?
These questions are not limited to AI labs and governments. They also concern those of us who use and build with AI.
Are we still using AI as a tool, or are we beginning to delegate decisions we can no longer properly verify?
Which skills and judgements do we deliberately want to retain?
And would we accept the same use of AI if we were not the ones controlling it—if an employer, public authority, insurance company, or platform used it to evaluate us?
Perhaps the central question is therefore not simply whether technological progress should be slowed down. It is whether our institutions, organisations, and individual judgement can develop quickly enough to govern it responsibly.
And if they cannot, what should we accelerate: the technology, our ability to govern it—or our willingness to set limits?
I do not think there is an easy answer. But waiting until the answer becomes obvious may mean waiting until meaningful choices have already disappeared.
Cover image created with ChatGPT using AI image generation.
Top comments (0)