NVIDIA has announced the formation of the Open Secure AI Alliance (OSAA), a coalition of 37 founding organisations dedicated to improving the security of artificial intelligence through open collaboration. Alongside the alliance, NVIDIA has also open-sourced its NOOA (NVIDIA Object-Oriented Agents) framework, a Python-based platform designed to help developers build, test, and secure AI agents more effectively.
The new alliance brings together major technology companies, security vendors, cloud providers, open-source organisations, and AI developers to create and share open technologies, tools, and best practices for securing AI systems. The initiative comes as AI agents become increasingly capable and widely deployed, raising concerns over vulnerabilities, misuse, and cyberattacks targeting AI infrastructure.
According to NVIDIA, the alliance aims to accelerate the development of open AI security tools by encouraging members to contribute software, research, engineering expertise, and security frameworks. Rather than relying on proprietary approaches, the group believes that transparent, community-driven security solutions will help organisations identify and fix AI vulnerabilities faster.
One of the biggest announcements is NVIDIA’s decision to release the NOOA (NVIDIA Object-Oriented Agents) framework as open source. NOOA introduces a programming model that treats AI agents as native Python objects, allowing developers to manage prompts, tools, memory, workflows, and agent behaviour using familiar Python classes and methods. This approach simplifies development while making AI systems easier to test, debug, audit, and secure.
NOOA also includes built-in support for tracing agent actions, validating model outputs, managing long-term memory, and integrating with security testing frameworks. NVIDIA says these capabilities will enable developers to build more reliable AI applications while reducing risks associated with autonomous agents.
The Open Secure AI Alliance includes organisations from across the technology ecosystem. Founding members include companies such as Adobe, Cisco, CrowdStrike, Dell Technologies, Hugging Face, Hewlett Packard Enterprise (HPE), IBM, Microsoft, Palantir, Siemens, SpaceX, and the Linux Foundation, among others. Several members are contributing their own open-source security technologies, including Microsoft’s MDASH scanning framework, HPE’s workload identity technologies, and Hugging Face’s Safetensors ecosystem.
The announcement follows growing industry debate over whether advanced AI models should remain open or closed. NVIDIA and its partners argue that open AI technologies are essential for cybersecurity because defenders need unrestricted access to understand, evaluate, and protect increasingly sophisticated AI systems. The alliance positions openness as a security advantage rather than a liability.
NVIDIA says the alliance will continue expanding its ecosystem by releasing additional open-source tools, security benchmarks, evaluation frameworks, and collaborative research projects designed to strengthen the safety and resilience of AI systems across the industry.
