The CEOs of Nvidia, OpenAI and Anthropic converged at Salesforce’s Dreamforce conference in San Francisco on Tuesday, bringing one of the technology industry’s most consequential debates directly onto its biggest enterprise stage.
Anthropic CEO Dario Amodei and Nvidia CEO Jensen Huang joined Salesforce CEO Marc Benioff during the opening keynote. Hours later, OpenAI CEO Sam Altman sat down with Benioff for an afternoon fireside chat.
Their appearances came days after Amodei urged the industry to slow the advancement of frontier artificial intelligence, a position Altman subsequently supported—and Huang forcefully challenged.
The disagreement gave Dreamforce unique prominence this year. While the companies building the world’s most powerful AI systems debated how quickly the frontier should move, Salesforce focused on a more immediate problem: helping businesses safely absorb the capabilities already arriving.
Three Views on AI Safety
Amodei initiated the debate over the weekend by calling for a coordinated effort to slow improvements in frontier models. He proposed independent safety evaluations, common standards among democratic countries and, ultimately, international cooperation around the most powerful systems.
At Dreamforce, the Anthropic CEO said developers should begin by improving their own safety practices before pursuing industry-wide and international standards.
“Let’s recommit to transparency. Let’s invest more in safety,” Amodei said. “Then let’s organize the rest of the industry and say, what can we do to set standards for everyone? And then finally, there should be an international component to it.”
Amodei said Anthropic has committed to the first part of that plan and intends to begin a dialogue with the rest of the industry about broader coordination.
Altman backed the central premise of Amodei’s proposal before arriving at Dreamforce, agreeing that AI companies need to “pace the frontier” and supporting the use of independent evaluators.
During his conversation with Benioff, Altman did not identify a particular release or development decision that OpenAI would delay. Instead, he argued that AI companies must act responsibly regardless of what their competitors do.
“Be responsible no matter what,” Altman said. “There should be no qualifier on that.”
The OpenAI CEO said public trust depends on companies making dependable decisions without treating commercial or geopolitical competition as an excuse.
Huang offered the clearest counterargument at the All-In Summit in Los Angeles on Monday. The Nvidia CEO challenged predictions of catastrophic AI outcomes and argued that companies concerned about moving too quickly could voluntarily slow their own development.
During Huang’s appearance, President Donald Trump called him onstage and dismissed fears of an AI takeover. Trump has argued that restrictions could weaken the United States in its competition with China.
Huang carried that argument into Dreamforce, where he agreed that safety must remain a priority but rejected the need for an industry-wide slowdown or additional regulation.
“Safety is paramount. In a lot of ways, it’s job one,” Huang said. “However, safety is an engineering problem.”
Huang said companies should rigorously test AI systems and decline to release products when they are not confident in their safety. But he called the perceived trade-off between innovation and safety “a false choice,” arguing that companies can move quickly while pausing when necessary.
Salesforce Bridges the Gap
The dispute arrived as Salesforce introduced AIforce, a new interface layer designed to connect AI tools with the data, permissions, workflows and business rules held inside its platform.
Benioff acknowledged that AI adoption remains uneven. Employees may already use products such as ChatGPT or Claude in their personal lives, he said, only to return to workplaces that have yet to undergo the same transformation.
But simply giving businesses access to increasingly capable AI systems will not be enough.
“Models alone cannot run the enterprise,” Benioff said. “Models alone are not going to show us what’s possible. Models also are very probabilistic.”
Salesforce is pitching AIforce as a bridge between that probabilistic world and the more deterministic information held inside corporate systems.
The technology lets employees and AI agents use Salesforce without navigating traditional layouts, dashboards and fields. Users can ask questions, update records, trigger workflows and generate customized interfaces by describing what they need.
AIforce is launching through Claudeforce, Slackforce and Agentforce Coworker, allowing employees to work with Salesforce information from Claude, Slack or Salesforce itself. The company says those experiences remain governed by existing permissions, security policies and business rules.
Altman said enterprise AI is now moving beyond chatbots and coding assistants toward systems that continuously understand an employee’s work and act proactively.
“The AI is running for you all of the time, understanding your job, what you need to be more successful, rendering new code for you, proactively watching your Slack or email,” Altman said. “I think we are on the precipice of this third phase.”
Salesforce occupies a different position from the frontier laboratories. Its opportunity lies less in building the most powerful general-purpose model than in helping companies connect models and agents to trusted corporate information.
That makes the gap between the speed of AI development and the speed of enterprise adoption especially consequential. More measured frontier releases could give businesses additional time to test systems, establish governance and train employees. Slower progress could also delay the productivity gains those businesses have been promised.
Amodei stressed that pacing the frontier would not mean freezing AI progress or discouraging adoption. Even if the technology stopped improving, he estimated that businesses are using only 5% or 10% of its potential value.
The comments exposed an important distinction: AI companies could pace the development of more powerful models while businesses accelerate their adoption of capabilities that already exist. Huang, however, questioned whether coordinated intervention is necessary at all.
Dreamforce remained primarily a showcase for Salesforce’s products and partnerships, but the appearances by Amodei, Huang and Altman put a larger question at the centre of its opening day: not only how quickly AI can advance, but how quickly it should—and who should set the pace.
Leave a Reply