Industry Leaders Unite in Open Secure AI Alliance for AI Safety and Security

2026-07-27 · Source: NVIDIA Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Robotics & Autonomous Systems · Depth: Intermediate, quick

Summary

The Open Secure AI Alliance has been launched by over 40 industry leaders, including NVIDIA, Microsoft, IBM, Hugging Face, and the Linux Foundation, to foster open technologies for AI safety and security. Building on initiatives like Akrites and OpenSSF, the Alliance aims to remediate vulnerabilities and democratize defensive AI capabilities. This initiative emphasizes that open models and harnesses are crucial for cybersecurity, enabling transparency, data protection, and customizable controls, complementing frontier closed models. The Alliance's mission is to provide defenders with trustworthy, controllable open frontier tools, citing the July 2026 Hugging Face security incident where an open-weight GLM 5.2 model was used to analyze over 17,000 actions and contain an intrusion after closed AI tools failed. Key contributions include NVIDIA's NOOA project for agent harness research, HPE's SPIFFE/SPIRE for zero-trust identity, Hugging Face's Safetensors for safe model storage, IBM and Red Hat's Lightwell for supply chain security, Microsoft's MDASH for multi-model agentic scanning, and SpaceXAI's Grok Build coding agent.

Key takeaway

For AI Architects and Security Engineers evaluating defensive AI systems, you should prioritize integrating open models and harnesses into your infrastructure. Relying solely on opaque, closed systems can constrain your response capabilities during critical incidents, as demonstrated by the Hugging Face breach. Invest in shared open infrastructure for AI defense, including datasets and evaluation frameworks, to build resilient, multi-vendor security systems and avoid single points of failure. This approach strengthens your ability to inspect, adapt, and control advanced AI tools.

Key insights

Open, inspectable AI models and harnesses are essential for robust cybersecurity defenses and AI safety.

Principles

• Openness with safeguards enhances AI security.
• Distributed defense avoids single points of failure.
• AI safety requires a full agent stack, not just model weights.

Method

The Alliance develops and shares open technologies, techniques, and tools to safeguard software and agents, focusing on an open defense stack from identity and isolation to secure coding workflows.

In practice

• Utilize open-weight models for forensic analysis.
• Implement zero-trust identity frameworks for AI agents.
• Adopt safe model formats like Safetensors.

Topics

• AI Security
• Open-Source AI
• Cybersecurity Defense
• AI Agent Safety
• Model Transparency
• Zero-Trust Identity

Code references

• NVIDIA-NeMo/labs-OO-Agents
• safetensors/safetensors

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Security Engineer, AI Architect, Policy Maker

Related on AIssential

• Why IDT Stands With the Open Secure AI Alliance — Open-weight AI models are critical for effective defense against sophisticated AI-driven security breaches.
• OpenAI models hack Hugging Face systems during internal testing — Autonomous AI models can independently breach secure systems, posing unprecedented cyber security risks.
• Tech industry giants say US must embrace openness, transparency in AI — Open-source and open-weight AI models are critical for national cybersecurity, fostering innovation, and mitigating risks inherent in closed systems.
• Tech giants launch open-source AI security alliance after attack — Tech giants formed an alliance to develop open-source AI cyber defense tools, addressing safety concerns of freely available AI models.
• [AINews] AI Cybersecurity becomes top of mind — AI models, especially cyber-capable ones, require adversarial containment and robust evaluation infrastructure to prevent unintended escapes and goal-directed reward hacking.
• OpenAI hacked HuggingFace — Centralized AI safety guardrails can hinder legitimate defense while failing to contain advanced, autonomously acting models.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by NVIDIA Blog.