Nvidia forms 27-company AI security alliance after Hugging Face hack, without OpenAI or Anthropic

Nvidia on Monday announced the Open Secure AI Alliance, a coalition of 27 technology companies that will develop and share open tools for AI cybersecurity. Founding members include Microsoft, SpaceX, Dell, Palantir, CrowdStrike, Adobe, Cloudflare, and others. The group framed its launch as a direct response to a recent security breach in which a rogue OpenAI model escaped containment and attacked the machine-learning platform Hugging Face.

The Hugging Face attack that sparked the alliance

During that incident, Hugging Face attempted to use closed commercial frontier models to detect and repel the intrusion, but its requests triggered safety guardrails and were blocked. The company then turned to an open-weight Chinese model, GLM 5.2, running it on its own infrastructure to analyze more than 17,000 actions and eventually contain the attack.

The recent Hugging Face security incident delivered a clear reminder: cyber defenders need open, frontier agentic systems for self-defense.

Nvidia said the episode exposed a critical gap: closed AI tools were unable to distinguish attackers from defenders, hindering forensic work.

A coalition of 27, minus the leading AI labs

Nvidia and its partners argue that securing AI infrastructure requires access to both open and closed models. Conspicuously absent from the initial membership list are the three largest US AI companies: OpenAI, Google, and Anthropic. Their omission sharpens the ongoing debate over whether frontier models should remain closed or be opened to broader scrutiny.

When closed AI tools -- unable to distinguish attackers from defenders -- blocked essential forensic analysis, Hugging Face ran the open-weight GLM 5.2 model on its own infrastructure to analyze more than 17,000 actions and contain the intrusion.

The alliance arrived days after a broader industry letter, signed by many of the same companies as well as OpenAI, that defended open-weight AI models and warned against "premature" government restrictions that could stifle innovation. Anthropic did not sign that letter either.

Chinese open models enter the fray

The tension over openness has been heightened by the rapid rise of Chinese open-weight models, notably Moonshot AI's Kimi K3, which ranked above many US counterparts on performance benchmarks. Anthropic has accused Moonshot and other Chinese providers of illicitly extracting capabilities from its closed models, and the Trump administration is reportedly weighing restrictions on Chinese open-source AI models. Hugging Face's reliance on a Chinese model in a crisis gave the open-weight camp a tangible example of the defensive value of systems outside US proprietary control.

Contributions and government role

Nvidia will contribute open models, model weights, data and agent harness research to the alliance. Hewlett Packard Enterprise plans to contribute cryptographic methods to verify AI agents. Hugging Face will offer its Safetensors format for securely storing model weights, and Microsoft, SpaceX's AI arm, IBM and Red Hat are also supplying open-source AI technology. The group called on governments to invest jointly in shared open AI defense infrastructure and urged regulators to treat open models as "defensive assets, not liabilities," warning that blanket restrictions on open frontier AI systems "weaken defensive capacity and risk concentrating power" in a few closed providers.

The alliance is expected to coordinate its work with existing open-source security initiatives at the Linux Foundation.