← Portada | Mercado

Anthropic CEO Dario Amodei Urges Slower Pace of Frontier AI Capability Gains

2026-09-13 · Fuente:Unite AI p3

Anthropic CEO Dario Amodei Urges Slower Pace of Frontier AI Capability Gains

In a September 12 essay, Anthropic's chief executive calls for deliberate pacing of model progress and pledges permanent employee-level access for outside evaluators.

Anthropic chief executive Dario Amodei has called on the artificial intelligence industry to deliberately restrain how quickly frontier models gain capability. The argument anchors a new essay, We Must Pace the Frontier, published on September 12, 2026. Alongside the essay, Amodei announced on his X account that Anthropic is unilaterally committing to give third-party evaluators permanent, employee-level access to its systems.

The essay's central claim is that the pace of model capability improvement must come down. "We must slow the pace at which we improve the capabilities of AI models," Amodei wrote. He added that progress will still seem fast, a caveat that frames the proposal as a moderation of the current trajectory rather than a stop to development.

The access pledge announced on X gives the essay a concrete counterpart. By offering outside evaluators permanent, employee-level access to its systems — and describing the commitment as unilateral — Anthropic ties its call for slower capability scaling to a standing transparency measure. Independent scrutiny of frontier systems, in this framing, becomes a mechanism that can accompany and verify deliberate pacing rather than a one-off audit.

For an industry racing to scale compute and capability, the essay injects a prominent voice for deliberate pacing from inside one of the leading frontier labs. The pairing of a slowdown argument with a standing access commitment gives Anthropic's position a testable component: outside evaluators will hold permanent, employee-level visibility into its systems. Whether other developers adopt a comparable model, and whether slower capability gains can hold against competitive pressure, now stand among the open questions raised by the September 12 publication.

Etiquetas: Anthropic · Dario Amodei · AI Safety · Frontier Models · Model Evaluation


Lecturas relacionadas


萬安算交所 AIXX · AIXX · Exchange verificado · Ir al mercado