NVIDIA and tech firms launch Open Secure AI Alliance for AI security
Written by Rebecca Uffindell Thu 30 Jul 2026
NVIDIA and a coalition of cloud, cybersecurity, enterprise software and open-source organisations have launched the Open Secure AI Alliance, an initiative focused on open tools for AI safety and security.
Inaugural partners include Microsoft, IBM, Mistral, Cloudflare, CrowdStrike, Cisco, HPE, Hugging Face, Red Hat, Salesforce, SAP, Siemens, Palo Alto Networks, Databricks, Dell Technologies, GitHub and Zscaler, alongside dozens of other companies and foundations.
The group says defenders need systems they can inspect, adapt and run on their own infrastructure, particularly as agentic AI moves further into production.
Open Tools Become Part of AI Security Planning
The scale of the alliance makes the announcement more significant than a single open-source project.
It brings together organisations spanning cloud infrastructure, cybersecurity, enterprise software, AI research and open-source development around a shared objective: giving defenders access to tools they can examine, test and improve.
The alliance builds on work from the Linux Foundation’s Akrites initiative and the Open Source Security Foundation community, with an emphasis on vulnerability remediation, disclosure and shared defensive infrastructure.
NVIDIA said the goal is not to replace closed models entirely. The company argued that defenders need both open and closed systems, depending on the task.
AI Security Depends on the Full Agent Stack
The alliance draws a distinction between a model and an operational AI agent.
As NVIDIA put it, “An AI agent isn’t just a language model. It is a complex system built from models, harnesses and guardrails.”
That difference matters for enterprise security teams. A model may generate outputs, but production AI systems also depend on the controls that determine what they can access, which actions they can take and how those actions are monitored.
The alliance has argued that real AI safety and security depend on the full agent stack — identity, permissions, harnesses, guardrails, logs, and evaluation — not just on whether model weights are open or closed.
For enterprises, that alters the focus toward traceability, auditability, and operational control once AI systems are connected to business workflows.
Cybersecurity is a Central Use Case
The alliance positions cybersecurity as one of the strongest use cases for open AI infrastructure.
Its argument is that defenders need technologies they can study and run locally, particularly when responding to active incidents or protecting sensitive environments.
NVIDIA said open models and open harnesses are important because they can increase transparency for defenders, protect data, and complement closed models with localised controls.
The announcement also cited a recent Hugging Face security incident. According to NVIDIA, Hugging Face used an open-weight model running on its own infrastructure to analyse more than 17,000 actions during forensic work after closed AI tools blocked parts of the analysis because they could not distinguish defensive activity from malicious behaviour.
The example underlines the practical case for AI systems that security teams can control directly when response speed and operational context matter.
Members Are Contributing Across the Stack
NVIDIA said it will contribute open models, model weights, data and agent-harness research to the alliance.
Its new NVIDIA Labs Object-Oriented Agent project, known as NOOA, is intended to make agent behaviour easier to test, trace, audit, and govern.
Other members are contributing tools across different layers of the AI security stack.
HPE is supporting zero-trust identity standards through SPIFFE and SPIRE. Hugging Face has contributed Safetensors, a model format designed to reduce the risk of remote code execution. IBM and Red Hat are working on digitally signed patches through Lightwell, while Microsoft’s MDASH project uses multiple specialised AI agents to identify and validate exploitable vulnerabilities.
Together, the work shows how AI security is becoming a systems issue rather than a model issue.
Policymakers Are Part of the Audience
The alliance is also making a case to policymakers.
It says open models, harnesses, and security tools should be treated as defensive assets rather than liabilities, warning that blanket restrictions could weaken cyber defence and concentrate risk in a small number of closed providers.
That creates a more complex policy question. The issue is no longer whether open or closed systems are safer in the abstract. It is whether defenders have access to the transparency, control, and deployment flexibility needed to secure AI systems in real operating environments.
AI Security Becomes Shared Infrastructure
The launch of the Open Secure AI Alliance reflects a growing view that defenders need more than capable models.
They need inspectable systems, shared standards and tools that can be deployed across different infrastructures without creating new single points of failure.
The breadth of the coalition also matters. With support from major cloud, cybersecurity, enterprise software and open-source organisations, AI security is being treated as a shared infrastructure challenge.
As agentic AI moves further into production, the organisations best placed to manage it may be those able to observe and govern the full system around the model.
Written by Rebecca Uffindell Thu 30 Jul 2026