-
NVIDIA is working with more than 120 members of the Open Secure AI Alliance to help make AI security more transparent, practical, and collaborative. The Alliance is contributing open tools across the AI security stack—from identity and permissions to agent harnesses, runtime guardrails, observability, evaluation and recovery.
-
The key idea: an AI agent is more than a model. The surrounding system determines what the agent can access, which tools it can use, how its work is orchestrated, what gets logged, and how it recovers from failures. If you want to secure agentic workflows, those layers need to be inspectable and testable too.
-
NVIDIA’s contributions include the NVIDIA Labs Object-Oriented Agent (NOOA) research harness, available on GitHub. NOOA makes agent behavior easier to test, trace, audit, and govern. It is a research harness—not a replacement for the framework you already use—but a useful surface for examining how agents behave and for sharing what works across the community.
-
NVIDIA OpenShell adds runtime controls that restrict what an agent can see, touch, and do. NVIDIA’s verified agent skills extend that trust to the capability layer: skills are cataloged, scanned for risks such as prompt injection and tool poisoning, cryptographically signed and documented with skill cards so developers can understand what an agent capability does and where it came from.
-
Industry leaders in the Open Secure AI Alliance also proposed Shared AI Findings Exchange (SAFE) guidelines. The goal is to help the ecosystem confidentially analyze AI incidents and near misses, identify recurring control failures, and share evidence-based recommendations that other teams can inspect, adapt and improve.
-
Open models, open tools, and open research do not eliminate security work, but they give more developers a way to participate in it. If you are building agent systems, explore the contributions, test them in your own stack and share what you learn.
Resources: