Nvidia has formed the Open Secure AI Alliance (OSAA) alongside over 30 global companies to create open-source tools for AI security. This collaboration addresses growing risks by building frameworks to protect AI systems and infrastructure. The initiative aims to enhance safety while maintaining the benefits of open AI development.
Detailed Coverage
Nvidia Corporation has taken a leading role in addressing the cybersecurity challenges of the artificial intelligence era by launching the Open Secure AI Alliance, or OSAA. This new coalition brings together more than 30 major technology, cybersecurity, and enterprise software companies. The group's primary goal is to develop open-source frameworks and tools that help organizations better protect their AI models and related infrastructure.
As AI systems become more autonomous, they also become more complex, creating new potential entry points for cyber threats. By using an open-source approach, the alliance intends to mirror the successful collaborative models seen in general software development, such as those overseen by the Linux Foundation. Instead of creating entirely new standards from scratch, OSAA plans to build on existing security efforts to speed up the adoption of safety measures.
The alliance will focus on several high-priority tasks, including the security of AI agents, foundational models, and the software supply chains that support them. Members are also working on advanced red-teaming, a practice where experts test systems by attempting to break them, to identify vulnerabilities before they can be exploited. Other focus areas include creating standardized benchmarks and datasets that companies can use to rigorously test their own AI security protocols.
Founding members include major global players such as Microsoft, Cisco, IBM, Adobe, Dell Technologies, CrowdStrike, Red Hat, Palo Alto Networks, Snowflake, ServiceNow, Hugging Face, Palantir, and SpaceX AI. These participants have agreed to contribute specific technologies and expertise to the project. For example, Nvidia is providing its NOOA project, while Hugging Face is contributing its Safetensors format. Other contributions range from zero-trust identity frameworks to agentic scanning tools developed by Microsoft.
Nvidia has defended the role of open models in this strategy, noting that open approaches allow for better transparency and localized control. While acknowledging that bad actors could theoretically misuse open-source AI, the company argues that restricting access is not the solution. Instead, the alliance advocates for a combination of transparency, clear rules against misuse, and rapid responses to any security gaps. The success of this initiative will likely depend on how quickly these firms can move from setting up the coalition to releasing functional tools that developers and enterprises can realistically integrate into their existing security workflows.