Amodei Calls for Slower AI Progress; Huang Rejects New Rules at Dreamforce

The executives agree companies must act safely, but offered sharply different tests for when AI development should slow.

By 2 min read
Amodei Calls for Slower AI Progress; Huang Rejects New Rules at Dreamforce
Amodei Calls for Slower AI Progress; Huang Rejects New Rules at Dreamforce

Listen to this story

The audio brief

About 1:20
0:001:20
Read transcript
At Salesforce’s Dreamforce keynote, two of AI’s most prominent executives offered sharply different answers to one question: when should the industry slow down? Dario Amodei, CEO of Anthropic, argued that companies can deliberately moderate the pace of capability gains without giving up commercial competition or U.S. leadership. His appearance came days after he published an essay calling for slower model development. The proposal is not to stop progress, but to make its speed an explicit industry choice before an immediate crisis appears. Jensen Huang, Nvidia’s CEO, rejected that approach. Speaking after Amodei, Huang called the conflict between moving quickly and pacing development a “false choice.” In his view, companies should run as fast as they can, while keeping safety in focus. A pause should come when developers believe a product may be unsafe, or when the company itself feels out of control. Huang also said existing market forces are sufficient and opposed additional AI laws or regulations. That puts the executives on opposite sides of a governance threshold: Amodei favors preventive limits on capability growth, while Huang puts the decision in the hands of developers after a risk becomes visible. The unresolved issue is who gets to define that moment—companies, markets, or policymakers—and whether waiting for a warning is enough.

Story brief

3 key points

At Salesforce’s September 15 Dreamforce keynote, the disagreement shifted from abstract safety rhetoric to who gets to set an AI capability threshold. Dario Amodei argued that developers can intentionally moderate gains while preserving competition and U.S. leadership. Jensen Huang instead backed maximum speed, with pauses only when a product appears unsafe or a company feels out of control, and rejected new AI...

  1. 01

    Amodei’s position followed an essay published days earlier calling for slower model development.

  2. 02

    Huang described the conflict between speed and pacing as a “false choice.”

  3. 03

    Huang said existing market forces are sufficient and opposed additional AI laws or regulations.

A debate over who should set the speed of AI development played out onstage at Salesforce’s Dreamforce. Anthropic CEO Dario Amodei argued for tempering capability growth, while Nvidia CEO Jensen Huang said safety and speed can coexist—and opposed new AI rules.

A slowdown proposal reaches the keynote stage

Amodei’s appearance came days after he published an essay urging the industry to slow model development. At the September 15 keynote, he returned to that position, arguing that companies could deliberately temper how quickly model capabilities improve.

His case was not that companies should abandon competition. Amodei said the industry could slow capability growth without giving up commercial advantage or U.S. leadership in AI. That makes the dispute less about whether progress continues than about whether its pace should be an explicit industry choice.

Huang’s case for company judgment

Huang, who appeared after Amodei, called the supposed choice between moving quickly and pacing development a “false choice.” He said speed and safety are not mutually exclusive, placing the responsibility for deciding when to stop with model developers themselves.

Run as fast as you can. But if you feel at any given point in time the company’s out of control or the product’s not going to be safe, take a pause and make sure you get it right.

Huang also said existing market forces are sufficient and that the AI industry does not need new laws or regulations, according to CNBC. That position puts him at odds with a framework based on deliberately slowing capability gains before a company identifies an immediate problem.

The unresolved decision

The Dreamforce exchange exposed two different approaches to restraint. Amodei wants the industry to make room for safety by moderating capability growth; Huang’s approach relies on developers to recognize danger and pause when it appears. Whether companies, policymakers, or markets should decide that threshold remains the central divide.

Sources

  1. cnbc.comNvidia and Anthropic CEOs diverge on AI safety at Dreamforce: Huang says fast vs. pacing is 'false choice'

Loading discussion...