Day one of Dreamforce put a familiar question back at the centre of the AI debate: how fast should the industry move, and how much responsibility should developers bear for ensuring the technology is safe?
Dario Amodei, co-founder and CEO of Anthropic; Jensen Huang, founder and CEO of Nvidia; Sam Altman, CEO of OpenAI; and Salesforce chairman and CEO Marc Benioff offered four distinct takes on the question — days after it burst into the mainstream, following Amodei’s essay over the weekend urging the industry to slow the pace at which it improves model capabilities. That call drew public support from Altman, Elon Musk and Google DeepMind chair Demis Hassabis, even as US President Donald Trump dismissed fears of AI destroying humanity as a “hoax” and critics such as former White House AI czar David Sacks accused Anthropic and OpenAI of seeking to entrench a “duopoly”.
Appearing first during Benioff’s keynote before some 12,000 people at Moscone Center, Amodei was consistent with his essay’s argument, saying the industry should respond to safety concerns by first examining its own record rather than pointing fingers at competitors. Reaching for an automobile analogy, he asked how a carmaker ought to respond to a safety incident caused by a brake defect.
“You may not have had this big, high-profile incident (yet), but I’m sure we’re not perfect,” Amodei said, calling on companies to improve their practices, recommit to transparency and invest more in safety.
He also called for the industry to work together to establish common standards, followed by an international component to the effort. The three-step approach, he said, was aimed at setting an example for the wider industry and recognising that companies could always improve — without sacrificing commercial advantage or America’s lead in AI.
Huang, who took the stage shortly after, offered a different emphasis, arguing that the industry should keep moving quickly while making safety integral to development. “Safety is paramount,” he said, describing it as fundamentally an engineering problem.
His prescription: build proper test environments, test products thoroughly, and hold back systems if companies are not confident about their functionality, capability or safety. He rejected the notion that innovation and safety involve a trade-off, calling the choice between speed and pacing a false one. Companies should “run as fast as they can”, he said, but pause when systems appear out of control or products are not safe. Market forces, rather than new laws or regulations, could provide the necessary discipline, he argued.
Altman, in a fireside chat with Benioff later in the day, aligned more closely with Amodei, saying it was imperative for safety and monitoring to come before capabilities, given the power of existing models. He took aim at conditional commitments — companies promising to be responsible only if rivals are — saying the public wants safety assured “no matter what”. “There should be no qualifier on that,” he said.
Benioff, playing host and referee, said the technology itself is neither inherently good nor bad. “It’s what you’re doing with the technology that matters,” he said, adding that AI companies “have to be responsible for what they’re building”.
He also urged stakeholders to remember the lessons of previous technology revolutions. Citing social media and its harmful impact on sections of society such as children, Benioff stressed that the mistakes of that era should not be repeated.
The debate carried commercial subtext too. Salesforce, which expanded its partnership with Anthropic last month to integrate Claude with its products, used the keynote to showcase its own AI push, with about 50,000 people expected at the three-day conference.
(The writer is in San Francisco at the invitation of Salesforce)