The Linux Foundation is taking over the Open Secure AI Alliance (OSAA) from Nvidia. The alliance, which is just over a month old, is working on open-source tools, shared standards, and best practices to help organizations consistently inspect, audit, and secureAI systems .
With a series of AI-driven security incidents still fresh in everyone’s memory, the OSAA aims to establish a definition of safe AI use from within the industry itself. The EU has already taken the lead and is influencing the rollout of the latest models, such as watermarks for Fable 5.1.
At the Linux Foundation, the initiative will find what the foundation calls a neutral home for collaboration on the security of AI systems. Nvidia is taking a step back to give the alliance a single, open home rather than having it steered by a single vendor.
What SAFE is expected to do
One of the concrete outcomes is expected to come from the Shared AI Findings Exchange, or SAFE for short. This will enable organizations to collectively learn from AI incidents and near-misses, and translate recurring errors into evidence-based protective measures.
In a blog post, Nvidia explains that SAFE will be a framework that allows companies to confidentially collect and analyze incidents, inform affected parties, identify recurring “patterns of failure” in control systems, and publish operational recommendations.
Building on Akrites
The work is somewhat reminiscent of Akrites, the Linux Foundation initiative designed to defend critical open-source software against AI-driven cyber threats. Its backers include AWS, Anthropic, Cisco, Google, IBM, Microsoft, Nvidia, OpenAI, Red Hat, Sonatype, Vodafone, and Zscaler. Akrites focuses on coordinated vulnerability remediation and responsible disclosure. OSAA and findings in SAFE will provide additional context from the modeling perspective.
According to Nvidia, the work of the OpenSSF community also forms a foundation for the alliance. Red Hat joined the partnership at its launch, calling it a strengthening of the open-source defense layer.
See also: Open secure AI alliance shares SAFE guidelines for AI incidents