NVIDIA and partners launch Open Secure AI Alliance for AI agent security

NVIDIA and a broad group of technology partners have launched the Open Secure AI Alliance, a new industry effort aimed at developing shared open tools for securing AI systems and agentic software. The company announced the alliance on July 27, describing it as a way to give defenders technologies they can inspect, adapt and run under their own control instead of relying only on closed systems.

The launch arrives as AI agents are moving from chat interfaces into software development, security operations and enterprise automation. NVIDIA frames the problem as bigger than model weights alone: an agent also depends on identity, permissions, harnesses, logs, guardrails and evaluation. In that stack, opaque tooling can make it harder for defenders to understand why a system acted, what data it touched, and whether a compromise has been contained.

What the alliance plans to build

According to NVIDIA, inaugural partners include Microsoft, IBM, Cisco, Cloudflare, CrowdStrike, Dell Technologies, Hugging Face, the Linux Foundation, Palo Alto Networks, Red Hat, Salesforce, ServiceNow, Siemens and others across cloud, security, enterprise software and open source communities. The stated goal is to develop and share open technologies, techniques and tools that help safeguard software and AI agents as the technology evolves.

NVIDIA is also contributing NVIDIA Labs Object-Oriented Agent, or NOOA, as an open source research framework. The company says NOOA is designed to help agent harnesses integrate with models in ways that make behavior easier to test, trace, audit and govern. The Linux Foundation separately said the framework is being provided for general use and positioned the alliance as an extension of open source security practices such as shared tooling, transparent review and coordinated vulnerability handling.

Why it matters

The announcement gives the open model debate a practical cybersecurity focus. Supporters of open weights argue that defenders need comparable capabilities to detect, simulate and respond to attacks, especially when sensitive investigations must run on infrastructure they control. NVIDIA cited a recent Hugging Face security incident as an example of why organizations may need inspectable tools during live response work.

The alliance still has important details to prove, including governance, release cadence, security review processes and how competing vendors will coordinate vulnerability disclosure. But the partner list and the first code contribution make this more than a policy statement. If the group can turn its launch into maintained tools, it could shape how enterprises evaluate AI agents: not just by model performance, but by whether the surrounding control layer can be tested, audited and improved in the open.

Sources

• blogs.nvidia.com
• www.linuxfoundation.org

Cover photo by Brett Sayles on Pexels, used under the Pexels License.