What happens when the people racing to build the most powerful AI systems start admitting that the race itself may be moving too quickly?
That is the bigger story behind the latest debate involving OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei.
Amodei has called for the frontier AI industry to slow its pace, arguing that safety systems and society need time to catch up with rapidly increasing capabilities. His proposal includes stronger external evaluation and greater coordination between AI companies and governments.
Altman’s agreement is significant because this is not an outside critic demanding that AI companies stop innovating. It is coming from inside the frontier-model race.
And that changes the conversation.
The real bottleneck may no longer be simply computing power, chips or training data. It could increasingly become verification.
A model becoming more capable is one thing. Knowing exactly what that model can do, when it may behave unexpectedly, how it can be misused, and whether existing safeguards actually work is another.
Recent concerns around autonomous AI agents, cyber operations and misuse have made that distinction harder to ignore.
So “slow down” does not necessarily mean “stop AI.”
It could mean: build the safety infrastructure at the same speed as the intelligence itself.
The next phase of the AI race may therefore be measured differently.
Not just by who releases the smartest model first.
But by who can prove that increasingly powerful models can be deployed responsibly.
That could become one of the most important competitive advantages in AI.
The AI race is accelerating. But perhaps the real race now is to make that acceleration controllable.
