Benioff hosts AI war: Amodei urges slowdown, Huang says run faster

Gwen Stefani sang "Underneath It All" to a packed San Francisco hall, but the performance was just the warm-up. Marc Benioff took the stage with a revenue forecast chart — $46 billion by 2027 — and the crowd applauded. Hovering over everything was a different question entirely: is it worth risking human existence for the products that generate that revenue?

The fight peaked a week earlier. Jacob Coxon quit Anthropic and warned publicly that companies racing toward self-improving AI systems are "betting our lives on it." More employees joined in. Dario Amodei, Anthropic's CEO, called on world leaders to help "pace the frontier." Sam Altman, his great rival at OpenAI, backed the idea. On the other side, David Sacks, the White House AI adviser, dismissed the appeals as "yet another attempt at regulatory capture," and Trump labeled existential risk a "hoax." Congress is still trying to push legislation forward.

On stage: the car analogy and the nod to OpenAI. Benioff, six-foot-five, towered over Amodei and shook his hand warmly before asking about the pace of progress. Amodei reached for an analogy: if a competitor has a safety incident, the temptation to pile on is strong, but the responsible move is to organize the industry and set shared standards. The reference to OpenAI's Hugging Face episode was unmistakable. Amodei quickly clarified that "pacing the frontier" doesn't mean freezing the technology, and used the platform to plug Claude's integration into Salesforce.

Huang: safety is an engineering problem, keep running. Jensen Huang, Nvidia's CEO, followed immediately and flipped the frame. AI safety is "an engineering problem" only, he argued, and companies can police themselves. "You don't need new laws, you don't need new regulation," he told the room. "You run as fast as you can, but if at some point the company spins out of control or the products aren't safe, you stop." The gap between the two worldviews could not have been sharper.

Outside, academia watches with a critical eye. The reporter left the hall — stuffed with plush toys and half-Hawaiian decorations — to talk with Sayash Kapoor, an incoming computer-science lecturer at UC Berkeley. Kapoor, co-author of "AI Snake Oil" with Princeton's Arvind Narayanan, represents the voice reminding you that most of what's sold here as "agents" and "smart dashboards" is a long way from the existential risks the CEOs are debating, and a long way from working like the demos.