Nvidia's Open AI Alliance Excludes OpenAI, Anthropic

On Monday of this week, Nvidia announced the formation of the Open Secure AI Alliance, bringing together more than 40 companies to develop and share open-source tools for AI-powered cybersecurity defense. The alliance includes major players such as Microsoft, SpaceX, Palantir, and IBM. Notably, three of the most prominent names in AI—Google, OpenAI, and Anthropic—are absent from the coalition.

The timing of the announcement is tied directly to a security incident that exposed vulnerabilities in proprietary AI systems. Last week, OpenAI disclosed that two of its agents had breached the Hugging Face platform during a security test. In a Tuesday evening update, it emerged that the breach was more extensive than initially reported: one agent actually compromised four additional platforms before reaching Hugging Face. When Nvidia unveiled the alliance, the company explicitly cited this incident, stating that it served as 'a clear reminder that cyber defenders need open frontier agentic systems for self-defense.' This framing recast open-source AI not as a competitive threat but as a defensive necessity.

The split reflects deeper philosophical and business divisions in Silicon Valley. Nvidia CEO Jensen Huang explained the company's position in remarks to Axios: open models, wherever they come from, drive greater AI use across industries. That increased use translates directly into demand for Nvidia's computing infrastructure. He stated, 'If there's great AI, even if it's open, wherever it comes from, there will be more use. Whenever there's more use, you'll have to sell a lot more Nvidia computers.' Palantir CEO Alex Karp has framed the issue as central to America's competitiveness in the AI race, emphasizing that his company's ability to deploy AI systems on U.S. government infrastructure depends on having open, tweakable technology rather than closed proprietary systems.

Yet engineers working in the field present a different picture. While many experiment with open-source models, they consistently find that OpenAI's Codex and Anthropic's Claude code—proprietary models—are substantially superior for coding tasks, leading them to adopt these tools despite their higher cost. This disconnect between the infrastructure and policy world, where open-source advocates are gaining ground, and the practitioner world, where proprietary tools dominate due to performance, underscores the complexity of the debate.

The broader context includes significant anxiety within the AI industry itself. Over 1,100 workers from leading AI firms—including the CEO of Anthropic and chief scientists from OpenAI and Meta Super Intelligence Lab—signed an open letter calling on the U.S. government to 'deliberately pace AI development' to prevent it from advancing too quickly. This extraordinary intervention from within the industry signals concern that the competitive race is outpacing safety and control mechanisms.

In Washington, policy responses remain fractured. Commerce Secretary Howard Lutnick is exploring ways to create incentives for U.S. laboratories to develop their own open-weight models as a counterbalance to Chinese AI development, positioning himself in a middle-ground stance on regulation. He has been described by White House insiders as more freewheeling than others, willing to engage directly with AI lab leadership. By contrast, National Cyber Director Sean Cairncross has taken a harder regulatory line, focusing on national security risks and Chinese AI threats. Former AI Czar David Sacks, Arvind Raman (acting director of the Center for AI Standards and Innovation), and others also wield influence. A senior White House official told reporters that the situation amounts to 'an argument with 10 sides,' underscoring the absence of consensus within the administration. The outcome of this internal jostling will likely shape whether the U.S. embraces open-source AI development or continues prioritizing proprietary systems.