What execs and politicians are saying about slowing down AI development

Anthropic CEO Dario Amodei’s Saturday essay “We Must Pace the Frontier” proposed three steps: embedded third-party evaluators, coordination among democratic-country frontier labs, and global government coordination “to the extent this is possible.” Anthropic unilaterally committed to the first step while stressing pacing is not a full halt to training.

Sam Altman agreed and backed independent evaluators plus a federal safety framework. Demis Hassabis called the direction right. Elon Musk simply said “Dario is right.” President Trump blasted the idea on Truth Social; VP JD Vance called industry begging for regulation a “Trojan horse.”

We’re treating this as the week’s primary policy tape — agreement at the lab layer, open conflict in Washington.

We’re reading a split screen. Lab chiefs are converging on language about slower-than-maximal progress, shared standards for misalignment and monitoring, and outside evaluators. Altman’s clarifying line matters: pacing does not mean stopping. That keeps the race alive while buying political cover.

Trump’s post is the counter-program — guardrails equal a “STRONG AND SMART” president, Anthropic cast as a sudden angel, and slowdown talk framed as a sick conspiracy that only helps China. Speaker Mike Johnson wants a meeting but rejects a moratorium for national-security reasons. Bernie Sanders and Pete Buttigieg land closer to pause-or-act urgency.

Useful transparency from embedded evaluators would be a real upgrade if labs honor Amodei’s first step. The downside if this stays quote theater is obvious: CEOs collect safety credit, regulators stall, and China remains the universal veto on any hard limit. I’m watching who actually seats auditors inside labs and whether Johnson’s promised sit-down produces anything sharper than vibes.

Nvidia’s Jensen Huang, on speakerphone with Trump at the All-In Summit, said America would make sure everybody wins the AI race. Altman also warned progress could go badly via loss of control to AI or excessive concentration of power.