The AI Slowdown Debate Crashed Salesforce's Party
By Maxwell Zeff | September 17, 2026 | Business
The Dreamforce conference became an unlikely battleground for the CEOs of OpenAI, Anthropic, and Nvidia to debate whether AI development should slow down.
In a year when AI regulation has moved from think pieces to draft legislation, Salesforce's annual Dreamforce conference was expected to spotlight the company's latest enterprise AI tools. Instead, the event turned into a high-stakes debate over whether the breakneck pace of AI development should be slowed—a question that has gained new urgency as models grow more capable and their economic impact deepens.
The panel, featuring OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, and Nvidia CEO Jensen Huang, was meant to explore the future of enterprise AI. But it quickly pivoted to existential risk and the responsibilities of companies pushing the frontier.
Amodei, who has long advocated for a cautious approach, warned that without coordinated global governance, the race to build ever-larger models could outpace our ability to ensure they remain safe. "We're not just building products; we're shaping the conditions of our own future," he said, calling for an international moratorium on training runs above a certain compute threshold until safety standards are agreed upon.
Altman pushed back, arguing that slowing down would cede leadership to less scrupulous actors and delay benefits like personalized medicine and climate modeling. "The answer isn't to stop—it's to build guardrails that scale with the technology," he said, pointing to OpenAI's recent work on interpretability and its participation in the Frontier Safety Framework.
Huang, whose company supplies the hardware underpinning most large models, took a pragmatic stance. "Compute is a tool. How we use it is a choice," he said. "But pretending we can put the genie back in the bottle is naive. We need to invest in safety research, not pretend we can halt progress."
The exchange reflected a broader industry schism that has only widened in 2026. On one side are those who believe AI is approaching a point where its risks—from disinformation to autonomous weapons—demand preemptive brakes. On the other are those who see such caution as a recipe for stagnation and a gift to competitors.
Salesforce, which has staked its own future on integrating AI across its CRM platform, found itself an awkward host. CEO Marc Benioff, who has publicly called for a "moonshot" approach to AI safety, tried to steer the conversation back to customer value but acknowledged the tension. "We're all trying to figure out how to do this right," he said during a later session. "But we can't let fear paralyze us."
Indeed, the debate at Dreamforce mirrored conversations happening in boardrooms and capitols worldwide. The EU's AI Act is now in its enforcement phase, and in the US, a bipartisan bill proposing a licensing regime for advanced AI models is gaining traction. Meanwhile, China has accelerated its own regulatory framework, adding a geopolitical dimension to the slowdown debate.
For attendees, the panel was a reminder that the AI industry's internal disagreements are no longer academic. They have real implications for investment, regulation, and public trust. As one enterprise architect put it, "I came here to learn about AI agents. Instead, I got a crash course in existential risk."
The party may have been crashed, but the conversation is far from over. If anything, Dreamforce 2026 made clear that the question of whether to slow down AI is no longer a fringe concern—it's center stage, and the industry's biggest players are only beginning to grapple with it.
via Wired AI