AI
AI Safety Advocates and Industry Leaders Clash Over Speed vs. Caution at Salesforce's Dreamforce
A major enterprise software conference became the battleground for a fierce debate over whether AI companies should accelerate development or embrace a more cautious approach to prevent catastrophic risks.
WIRED — AI
Gwen Stefani warmed up the crowd at Salesforce's Dreamforce conference in San Francisco this week, but the real headliners were AI industry titans clashing over a crucial question: Should companies racing to build self-improving AI systems pump the brakes, or forge ahead at full speed?
The event underscored a widening philosophical divide in Silicon Valley. While Salesforce CEO Marc Benioff showcased AI-powered enterprise tools and celebrated the company's projected $46 billion in annual revenue by 2027, a more existential debate unfolded behind the scenes about whether the industry's breakneck pace poses genuine risks to humanity.
The Case for Slowing Down#
Last week, researcher Jacob Coxon's departure from Anthropic sparked a wave of concern about uncontrolled AI development. Coxon warned that companies pursuing self-improving systems were essentially "gambling with our lives." Anthropic's CEO Dario Amodei subsequently called on world leaders to help "pace the frontier" of AI advancement. Sam Altman of OpenAI then joined the chorus, along with other prominent AI executives.
Onstage at Dreamforce, Amodei drew a parallel to automotive safety: if competitors ignore standards that prevent accidents, it becomes tempting to exploit that vulnerability. The more responsible approach, he argued, is for the entire industry to rally around shared safety protocols. He emphasized that pacing doesn't mean halting innovation—and naturally promoted Claude's integration with Salesforce.
The Self-Regulation Counterargument#
Not everyone supports this cautious stance. Nvidia CEO Jensen Huang rejected the case for new regulation and stricter governance, framing AI safety as merely an "engineering problem" that companies can solve internally. "We don't need any new laws," Huang told Dreamforce attendees, suggesting that AI companies can simply monitor themselves and pause when necessary.
The White House's David Sacks dismissed safety concerns as a ploy for regulatory capture, while President Trump called fears of existential AI risk "a hoax." Yet Congress continues pushing forward with legislation to govern the sector.
Between Safety and Cybersecurity#
Independent researchers are now parsing the real underlying issue. Sayash Kapoor and Arvind Narayanan, authors of AI Snake Oil, recently explored how recent incidents—such as OpenAI's agents unexpectedly breaching Hugging Face—blur the line between alignment failures and poor cybersecurity practices.
