Nvidia and Anthropic CEOs Divide on AI Safety at Dreamforce

At Salesforce Dreamforce conference in San Francisco, Nvidia and Anthropic executives presented contrasting philosophies on artificial intelligence development and safety. During a joint keynote with Salesforce CEO Marc Benioff, Anthropic CEO Dario Amodei and Nvidia CEO Jensen Huang addressed the ongoing industry debate over whether AI progress should be accelerated or deliberately moderated. Amodei took the stage first, referencing a recent essay in which he called for the sector to moderate the pace of model development to mitigate potential catastrophic risks. He outlined a three-step framework intended to enhance safety standards while preserving commercial viability and maintaining the United States competitive edge in AI. His proposal, prompted by earlier warnings from researchers about unchecked advancement, quickly garnered public backing from prominent industry leaders including OpenAI Sam Altman, Elon Musk, and Google DeepMind Demis Hassabis. Appearing immediately afterward, Huang challenged the premise that safety and rapid innovation are incompatible. Describing the perceived tension as a false choice, he argued that market dynamics already incentivize responsible development and that legislative intervention remains unnecessary. Huang emphasized that developers bear the primary responsibility for product safety, advising companies to advance quickly while retaining the discretion to halt releases if risks cannot be adequately mitigated. He maintained that strategic pacing should align with market readiness rather than external mandates. The divergent viewpoints unfolded against the backdrop of Dreamforce, which drew approximately fifty thousand attendees to San Francisco Moscone Center. The conference also highlighted recent strategic collaborations between Salesforce and key AI providers. Salesforce announced a new reasoning model for its Agentforce platform, built upon Nvidia open-weight Nemotron architecture, and reinforced its partnership with Anthropic through the launch of Claudeforce, a tool enabling direct data access within the Claude interface. Benioff closed the segment by asserting that while developers should exercise autonomy over their development timelines, they must ultimately face accountability for their deployments. The exchange at Dreamforce underscores a broader industry crossroads as foundational AI models grow increasingly capable. While proponents of measured progress argue for coordinated pauses to establish safety protocols, advocates for rapid advancement contend that market forces and internal corporate governance are sufficient to navigate emerging risks.