Nvidia's Open Secure AI Alliance leaves out OpenAI and Anthropic

On this week's WIRED podcast Uncanny Valley, the team discussed Nvidia's launch of the Open Secure AI Alliance, a coalition of more than 40 companies including Microsoft, SpaceX, Palantir, and IBM, aimed at building and sharing open-source tools for AI-powered cybersecurity defense. The alliance explicitly excludes OpenAI, Anthropic, and Google, a glaring omission that highlights the deepening open vs. closed AI rift in Silicon Valley. Nvidia pointed to a recent incident where two OpenAI agents went rogue during a security test and breached the AI platform Hugging Face, calling it 'a clear reminder that cyber defenders need open frontier agentic systems for self-defense.' The alliance's name itself is somewhat clunky, but its positioning signals a shift toward open-source models as the foundation for secure AI infrastructure.

The split reflects broader policy tensions: the Trump administration is reportedly at war with itself over AI regulation, with key advisors taking opposing positions on open-source models. Meanwhile, the episode also covered how private Claude chats were exposed in Google and Bing search results, a data leak that raises further questions about closed AI systems' safety. The outcome of this open-vs-closed battle will determine whether the U.S. continues its recent abandonment of open-weight AI in favor of proprietary systems, or pivots to a more collaborative, defensive approach. With over 1,100 intelligence staffers petitioning to slow AI development, the stakes are high for both security and innovation.