NVIDIA Launches Open Secure AI Alliance to Build Open-Source Defenses for AI Agents

A coalition of over 30 technology industry leaders, including NVIDIA, Microsoft, CrowdStrike, Cisco, IBM, Palo Alto Networks, and Red Hat, has launched the Open Secure AI Alliance.

The initiative aims to equip cyber defenders with transparent, community-driven AI security tools to counter increasingly sophisticated digital threats.

Building on the Linux Foundation’s Akrites project and the OpenSSF community, the alliance focuses on vulnerability remediation and open disclosure through inspectable, shared technology.

The alliance argues that defensive AI models should not remain locked inside opaque, proprietary systems. Open-weight models and inspectable evaluation harnesses allow security teams to study, adapt, and deploy security tools on their own infrastructure a vital capability when detection speed and operational transparency determine whether an intrusion is contained.

A recent incident at Hugging Face demonstrated this urgency. During a security breach, closed AI tools failed to distinguish forensic analysts from attackers, blocking investigation efforts.

Hugging Face’s incident response team switched to the open-weight GLM 5.2 model running locally, analyzing over 17,000 system actions to contain the intrusion.

As highlighted in NVIDIA’s official announcement, keeping security tooling transparent enables defenders to inspect and strengthen critical systems rather than relying on black-box providers.

Member organizations are contributing specific open-source components to build a modular “defense stack” for autonomous AI agents:

• NVIDIA NOOA: The NVIDIA Labs Object-Oriented Agent framework (available on GitHub) simplifies auditing, tracing, and testing agent behavior.
• Microsoft MDASH: A multi-model scanning harness that orchestrates AI agents to discover and validate exploitable software bugs.
• Hugging Face Safetensors: A secure file format designed to prevent remote code execution (RCE) risks within stored model weights.
• HPE (SPIFFE/SPIRE): Open identity framework contributions establishing zero-trust identity verification for AI workloads and agentic systems.
• IBM & Red Hat Lightwell: A supply-chain security project that secures open-source software via digitally signed patches.

This collaborative approach complements emerging open-source security suites built to streamline vulnerability management across enterprise environments.

Member Organization	Key Open-Source Contribution	Primary Defense Focus
NVIDIA	NOOA (Object-Oriented Agent)	Agent behavioral auditing & tracing
Microsoft	MDASH Harness	Automated bug discovery & validation
Hugging Face	Safetensors Format	Model weight RCE prevention
HPE	SPIFFE/SPIRE Extensions	Zero-trust agent identity verification
IBM / Red Hat	Lightwell	Digitally signed patch delivery

The coalition actively pushes back against claims that open AI models are inherently less secure than closed alternatives. While misuse risks exist, restricting access does not deter threat actors; it primarily limits defenders’ ability to inspect and harden infrastructure.

Rather than imposing broad restrictions on open frontier models, the group advocates for pairing openness with robust safeguards, including rigorous evaluations, anti-misuse policies, and rapid vulnerability remediation.

Furthermore, the alliance is engaging directly with regulators, urging policymakers to view open-source security tools as vital defensive assets.

Blanket restrictions risk concentrating power among a few closed ecosystem providers, ultimately weakening overall defense capabilities against sophisticated agentic AI tools deployed across the threat landscape.

“Transparency, not secrecy, provides the foundation for resilient AI security. By pooling resources across cloud, endpoint, and enterprise software sectors, the alliance ensures defenders retain the collaborative advantage.”

 Strengthen Your SOC by Accelerating Threat Detection & Rapid Investigations. -> Integrate ANY.RUN With Your SOC Now.