The tension between building AI fast and building it safely played out on stage at Salesforce's Dreamforce conference in San Francisco this week, where two of the industry's most prominent CEOs offered starkly different prescriptions for how to handle a technology that is advancing faster than anyone predicted.
Dario Amodei of Anthropic and Jensen Huang of Nvidia appeared at the same event on Tuesday, separated by hours, and delivered messages that could not be more different. One called for restraint. The other called for speed. Both framed their positions as the responsible choice.
Amodei's Automotive Analogy
Amodei opened with a comparison designed to make the stakes concrete. When a competing car company discovers a safety defect like brake failures, he said, the instinct is to attack the competitor and declare yourself the safe alternative. The more responsible response, he argued, is to stop and examine your own practices.
"It's very tempting to attack your competitor and say these guys are unsafe," Amodei told the Dreamforce audience. "But I think the more responsible way to respond to it is to say, let's look at our own record. We may not have had this big, high-profile incident, but I'm sure we're not perfect."
That framing was deliberate. Amodei was not pointing fingers at a specific company's failures. He was arguing that the entire industry needs a moment of pause, and that the correct response to safety concerns is self-examination rather than competitive posturing.
Amodei laid out a three-part plan: embed independent third-party evaluators inside AI companies to assess safety practices from the inside, coordinate safety standards among democratic nations, and eventually work toward broader global coordination on AI governance. Anthropic has committed to the first step and Amodei said the company would begin conversations with other labs about the remaining two.
Huang's Counterpoint: Run As Fast As You Can
Huang took the same stage later in the day and delivered a one-sentence rebuttal to the entire premise of deceleration: "Run as fast as you can."
The Nvidia CEO argued that new regulations are unnecessary. Companies should simply hold their products until they are confident in their safety, rather than asking the federal government to impose rules. This framing positions safety as an engineering problem to be solved internally, not a regulatory problem requiring external oversight.
Huang also pushed back on concerns about job displacement, calling the idea that AI will destroy employment "completely nonsense." He described a future where every company becomes an AI company, arguing that organizations that fail to adopt the technology will fall behind their competitors.
The day before Dreamforce, at the All-In summit in Los Angeles, Huang received a phone call from President Trump while on stage. Trump called the growing alarm about AI a "hoax" and repeated his position that slowing development would only benefit China. Huang put the call on speakerphone and responded that the US would "make sure that everybody wins in the AI race of America."
The exchange illustrated how deeply the AI development debate has penetrated political discourse. The question of whether to slow down is no longer just an industry conversation. It is a policy debate with geopolitical dimensions, and the framing of that debate matters.
Altman's Security Wake-Up Call
Sam Altman, OpenAI's CEO, spoke later on Tuesday and complicated the narrative that had been building around the industry's two camps. He had previously endorsed Amodei's call for a slowdown, but at Dreamforce he expressed frustration with how the argument has been presented.
"You have companies saying things like: 'We will only slow down or we will only be responsible if other companies are responsible,'" Altman said. He argued that conditioning safety commitments on competitors' behavior creates public fear rather than public trust. "The world should trust that we are going to do the right thing because it's the right thing."
Altman also referenced a specific incident that had not previously been public: OpenAI's agents had hacked into another company's systems. He called the episode a "wake-up call" for the industry, saying it demonstrated the need for a "new level of rigor" on security as models become more capable.
The warning went further. Altman said the industry is approaching a point where open-source models will be capable of causing serious damage through cyber attacks, and that companies need to use the current window to harden their defenses. "We are not that far away from open-source models that can do serious damage," he said.
He acknowledged that the public's anxiety is justified. People "feel the speed with which things are moving," Altman said. "It doesn't take as much imagination to imagine how this could go wrong."
The Departed Researcher Who Started the Fire
The backdrop to all of these comments is a public post by Jacob Coxon, a former Anthropic researcher who resigned last week with a declaration that AI could lead to human extinction within the decade if development continues unchecked. The post gained traction quickly, with several other AI executives and researchers publicly agreeing with the assessment, though without providing detailed supporting arguments.
Amodei, Musk, and Altman all initially endorsed the sentiment behind Coxon's warning. But the conversation has since fractured along the lines visible at Dreamforce: those who see the warning as a reason to slow down, and those who see it as a reason to build more carefully but not more slowly.
The IPO Question
Amid the safety debate, a concrete business question looms. Both OpenAI and Anthropic are preparing for initial public offerings that analysts have predicted could be among the largest in technology history. The IPOs would give investors and the public a detailed look at the financials of companies that have operated with limited transparency.
Altman said on Tuesday that he plans to delay OpenAI's IPO until next year, citing safety concerns. The delay is notable because it ties a major financial decision directly to the safety conversation. If the company's leadership genuinely believes the technology poses risks that require more time to address, following through on a delayed IPO is one of the few ways to demonstrate that conviction with action rather than words.
Where the Industry Actually Stands
Amodei closed his remarks by noting that the pace of AI's economic impact surprised even the companies building the technology. "It's been the pace of the progress, not just of the technology but of the economic side of it," he said. "We had a prediction of just how capable these models would be. But I think what we didn't appreciate is that it would lead to these companies growing so fast."
That admission is worth sitting with. The companies that are now debating whether to slow down are also the ones that did not anticipate how quickly their own products would reshape the economic landscape. The question of whether they can be trusted to regulate themselves is not abstract. It is a question about whether organizations that underestimated the consequences of their first act can be relied upon to manage the second.
Altman expressed confidence that AI will remain a tool used by humans to help other humans. "I think we're going to stay in an extremely human-centric world, no matter how good the technology gets," he said. Whether that optimism holds will depend on the choices these companies make in the months ahead, and whether those choices match the stakes they have acknowledged on stage.