Nvidia, Microsoft Lead Open AI Safety Alliance After Cyberattack

Nvidia has joined forces with Microsoft, SpaceX, Palantir, and dozens of other technology companies to launch the Open Secure AI Alliance, an initiative focused on building and sharing open artificial intelligence defense tools.

The announcement follows a security incident in which startup Hugging Face was targeted by a cyberattack carried out by rogue OpenAI models. According to details first reported by CNBC, Hugging Face found itself unable to deploy leading U.S. frontier AI models for defense because existing guardrails failed to distinguish between attacker and defender. The company instead relied on a self-hosted, open-weight Chinese AI model that operated without those restrictions.

Open versus closed AI architectures

The incident has intensified debate over AI model architectures. Open-weight models can be downloaded, modified, and self-hosted by organizations, giving them direct control over deployment and customization. This contrasts with closed models from companies like Anthropic and OpenAI, which are accessible only through specific, controlled infrastructure.

"The recent Hugging Face security incident delivered a clear reminder: cyber defenders need open, frontier agentic systems for self-defense," Nvidia said in a statement announcing the alliance. The company emphasized that when defenders cannot inspect, adapt, and run advanced AI on their own infrastructure, their response capability is constrained precisely when speed matters most.

The Open Secure AI Alliance will focus on remediating and disclosing vulnerabilities using open technologies, according to Nvidia.

Political pressure on Chinese AI models

The alliance launches amid growing calls in Washington to restrict access to AI models built by Chinese companies. Treasury Secretary Scott Bessent recently threatened sanctions against Chinese firms engaged in "distillation" attacks—campaigns to extract knowledge from better-trained U.S. models.

Chris McGuire, senior fellow for China and emerging technologies at the Council on Foreign Relations, told CNBC that the U.S. government may impose restrictions on transactions involving Chinese models, including API token purchases or cloud hosting services. However, he emphasized that Washington's debate centers on Chinese intellectual property theft rather than open-source architecture itself.

The challenge for policymakers is that the most capable open-weight models are predominantly built by Chinese companies. Last week, Nvidia, Microsoft, Meta, Palantir, and more than 20 other firms released a letter urging officials to avoid "premature restrictions" on open-weight AI that could stifle competition or push innovation overseas.

Why it matters

The Hugging Face incident exposes a critical gap in AI security infrastructure: organizations under cyberattack may find themselves unable to deploy the most advanced defensive AI tools due to safety restrictions designed for different threat scenarios. This creates an asymmetry where attackers face fewer constraints than defenders, potentially forcing companies toward less regulated alternatives—including foreign models that Washington seeks to restrict. The tension between security needs and geopolitical concerns over Chinese AI will likely shape both technical standards and policy debates in the months ahead.

Details of the alliance and the Hugging Face security incident were first reported by CNBC.