Breaking

Nvidia's Jensen Huang Rejects New AI Rules at Dreamforce as Safety Debate Intensifies

September 15, 2026 at 22:13 EDT

Policy & Regulation
AI Agents
Sales & Business

Nvidia CEO Jensen Huang used Salesforce's Dreamforce stage on September 15, 2026, to reject calls for a coordinated AI slowdown, arguing that safety should be handled through engineering, testing and release discipline rather than new regulation. The remarks sharpened a widening split over whether frontier AI companies should voluntarily slow capability gains while outside oversight catches up.

September 15, 2026 · Nvidia at Dreamforce

AI safety splits into two playbooks

Jensen Huang argued for engineering discipline and company-level release judgment. Dario Amodei is pushing a formal pacing plan with outside evaluators and coordinated standards.

steps in Amodei's pacing framework

6-12

months cited for agent-swarm risk

years of Salesforce CRM intelligence behind Koa

14+

industries represented in synthetic scenarios

Koa's benchmark claim: fewer CRM-action errors

Relative error index shown to scale; lower is better.

Leading models error index

Koa error index

3x fewer errors

Salesforce says Koa matched or exceeded leading CRM-action models in CRM Bench while making fewer wrong turns.

Huang's safety loop

Build test environments

→

Measure product safety

→

Release or pause

Engineering case

Supporters see testing, sandboxing, data boundaries and release gates as the practical path to safer AI products.

Oversight case

Critics argue company judgment may not cover systemic risks and that outside evaluators can make safety claims verifiable.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.