The author says Trump (“whoever wins AI wins,” with room for guardrails) and Anthropic’s Dario Amodei (“pace the frontier,” not halt training) are not a simple speed-vs-stop split. The useful rule is transition-specific: freeze a deployment, extra autonomy, or outside access when the evidence for that step fails, without shutting the whole lab.

An external evaluator can watch, report, delay, or block—those are different powers. Access without a pre-set decision owner is only transparency. Restart should need evidence that addresses the original failure, not time or a better benchmark. Reviewers should escalate without the team on the clock approving the finding.

Coordination could lock in incumbents; it needs a public purpose and oversight. Tighter chip and weight controls can sit beside a freeze on one dangerous permission so the U.S. lead is not treated as a reason to ignore a broken containment claim. Slower work is not, by itself, safer work.

Trump is right that the United States cannot casually surrender the AI race. Amodei is right that some capability transitions need brakes. The workable middle ground is to stop the permission the evidence no longer supports, not the entire field.

President Donald Trump and Anthropic CEO Dario Amodei appear to be offering opposite prescriptions for the AI race. Trump said Sunday that the United States must preserve its lead over China because “whoever wins AI wins,” while allowing that guardrails can still have a place. A day earlier, Amodei called for frontier developers to slow capability growth so that safety work has time to catch up.

The policy choice is less binary than those headlines suggest. In “We Must Pace the Frontier,” Amodei explicitly says pacing does not mean halting model training or technical progress. His proposal centers on embedded external evaluators, coordination among frontier companies and international cooperation. Trump, meanwhile, did not argue that every form of........

© Eurasia Review