What launched
NVIDIA announced the Open Secure AI Alliance, bringing together organizations in cloud computing, cybersecurity, open source, and AI. Building on the Linux Foundation's Akrites initiative and OpenSSF work, the alliance aims to develop shared technologies, tools, and evaluation methods for protecting software and AI agents.
Initial participants include NVIDIA, the Linux Foundation, Hugging Face, Microsoft, Cloudflare, IBM, Red Hat, Cisco, Databricks, Salesforce, SAP, NAVER, and SK Telecom.
Planned contributions
- NVIDIA plans to contribute open models, weights, data, and agent-harness research.
- The new open-source NOOA research framework is intended to make model–harness behavior easier to test, trace, audit, and govern.
- Members are working across agent identity and isolation, safer model formats, multi-model scanning, signed patches, and secure coding workflows.
Why it matters
Agent security is not only a property of model weights. Defenders must inspect permissions, tool calls, isolation, logs, guardrails, and evaluations across the full stack. Shared tools could let organizations test defenses without depending on one opaque vendor.
Limitations
This announcement establishes an alliance and a contribution roadmap. Maintenance, licensing, interoperability, and real-world detection performance must be evaluated project by project. Open models can also be stripped of safeguards or repurposed for attacks, so access controls, isolation, and responsible disclosure remain necessary.