Anthropic CEO Dario Amodei outlined a proposal to deliberately slow the pace of frontier AI development, according to a TechCrunch report published on September 12, 2026. The plan suggests coordinating release schedules and safety evaluations across leading labs rather than racing to deploy the most capable models first. OpenAI CEO Sam Altman has expressed similar sentiments, indicating rare alignment between the two competitors on the need to manage AI risk. Details remain vague, but the proposal implies voluntary commitments, shared benchmarks, and possibly regulatory backing. The core question is how to enforce pacing without ceding advantage to less cautious actors.
I read Amodei's plan as a pragmatic pause, not a surrender. Slowing the frontier is not about stopping AI; it is about making sure we understand what we are building before we ship it. That is how mature industries operate. We do not test new drugs on the open market. We do not fly untested planes. AI deserves the same discipline.
Critics will call this fearmongering. I call it foresight. The alternative is a race where speed trumps safety, and the winner is whoever cuts the most corners. That is a race we all lose. Pacing the frontier buys time for alignment research, red-teaming, and public debate. It also forces labs to compete on trust, not just raw capability. That is a healthier competition. Yes, there is a risk that bad actors ignore the pace. But a coordinated slowdown among leaders sets a norm. Norms matter. They shape what is acceptable. I would rather see a deliberate step back than a blind sprint forward. The future is not something we should stumble into. It is something we should design.