Nvidia’s AI Security Alliance Excludes OpenAI, Google, Anthropic
Nvidia has launched a new coalition of more than 30 major tech firms, including Microsoft, SpaceX, and the Linux Foundation, called the Open Secure AI Alliance. The aim of the group is to build and distribute open-source tools that improve safety and security against AI attacks.
As comprehensive as the group appears to be among household-name tech companies, though, there are notable exclusions: OpenAI, Google, and Anthropic, the world’s largest frontier-model developers.
Citing the recent automated attack on HuggingFace by OpenAI’s out-of-control AI system, the Nvidia coalition launch announcement suggested this was a clear indicator that more robust anti-AI security systems needed to be developed. Crucially, these systems need to be widely available rather than controlled by proprietary organizations.
To help develop this new security ecosystem, the coalition has brought in titans of infrastructure, software development, cybersecurity, and open source foundations. Standouts include Nvidia, Microsoft, SpaceX, Dell, Cloudflare, Cisco, Adobe, Red Hat, Salesforce, Siemens, SK Telecom, IBM, Hugging Face, and Capital One, among many others.
Credit: Nvidia
For its part, Nvidia will contribute open models, model weights, and data on new agent designs to help create better institutional security. Hugging Face has proposed “safetensors,” which would create a safe way to store model weights. IBM and Red Hat have put forward their Lightwell initiative, which offers AI-driven remediation capabilities to help fix flaws in software and platforms.
SpaceXAI and Microsoft have opened up their respective Grok Build and MDAS systems to help improve wider security systems, too.
The alliance is also pushing for policymakers and regulators around the world to take AI security more seriously, and to jointly invest in these new platforms and programs that can help protect against AI and agentic AI attacks.
OpenAI, Google, and Anthropic haven’t commented publicly on the alliance at the time of writing, though Anthropic has called for greater government control of new models and access, suggesting safety checks should be completed before any models are made publicly available—including those from overseas.
