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

Open-source software has long served as the bedrock of the modern global economy. From the intricate machinery of cloud computing and financial services to the essential infrastructure of telecommunications, manufacturing, and government operations, open-source projects provide a transparent, observable, and accessible foundation for innovation. Cybersecurity, in particular, has emerged as one of the primary beneficiaries of this collaborative ethos. Today, a massive coalition of industry titans is formalizing this commitment to collective defense through the launch of the Open Secure AI Alliance.

Building upon the foundational work of the Linux Foundation’s Akrites initiative and the OpenSSF community, this new alliance aims to standardize the remediation and disclosure of vulnerabilities within the AI ecosystem. As the United States and its global partners navigate the future of artificial intelligence, they face a pivotal decision: whether the security of our critical infrastructure will remain locked within opaque, proprietary systems or be constructed upon open models, harnesses, and tools that allow any defender to study, adapt, and deploy them.

The Case for Open Defensive Infrastructure

The prevailing consensus among the alliance members is that the world requires a hybrid approach, utilizing both closed and open models. However, when it comes to cybersecurity, openness is not merely a preference—it is a necessity. Open models and harnesses democratize defensive capabilities, foster transparency, and enable robust cyber defense without compromising sensitive data. By allowing for customizable, localized controls, these systems complement frontier closed models, providing a distributed, community-driven defense mechanism that eliminates single points of failure.

"The world needs both closed and open models. For cybersecurity, open models and open harnesses are essential because they democratize defensive capabilities, increase transparency for defenders, enable cyber defense while protecting data, and complement frontier closed models with customizable, localized controls."

While critics often argue that open models are inherently dangerous due to potential misuse, the alliance maintains that these risks are universal. Keeping model weights closed does not deter sophisticated, determined attackers. Instead, the most effective path forward involves pairing openness with rigorous safeguards, clear protocols against malicious behavior, and rapid, community-led remediation.

Lessons from the Frontlines: The Hugging Face Incident

The necessity of this alliance was underscored by a recent security incident at Hugging Face. When proprietary AI tools—unable to distinguish between malicious actors and legitimate security researchers—blocked essential forensic analysis, the team was forced to pivot. By deploying the open-weight GLM 5.2 model on their own infrastructure, they were able to analyze over 17,000 actions and successfully contain the intrusion.

This event served as a practical demonstration of a critical truth: when defenders are denied the ability to inspect, adapt, and operate advanced AI on their own terms, their response capabilities are severely constrained during high-stakes scenarios. To prevent future vulnerabilities, organizations across the globe require open, frontier-grade defensive tools that allow for a multi-vendor ecosystem, ensuring that no single entity becomes a point of failure.

A Massive Coalition of Industry Titans

The Open Secure AI Alliance represents a historic gathering of leaders from cloud computing, enterprise software, cybersecurity, and AI research. The inaugural partners include:

• Tech & Cloud Giants: NVIDIA, Microsoft, IBM, Dell Technologies, HPE, SAP, Salesforce, and Oracle.
• Cybersecurity Leaders: CrowdStrike, Palo Alto Networks, Fortinet, Cisco, Wiz, and Zscaler.
• AI & Research Innovators: Hugging Face, Databricks, Cohere, Mistral, Perplexity, and OpenAI-adjacent research labs.
• Infrastructure & Platforms: GitHub, Docker, Red Hat, Snowflake, Nutanix, and Cloudflare.

This diverse group is committed to developing and sharing the technologies, techniques, and tools necessary to safeguard software agents in the rapidly evolving AI landscape.

Building the Open Defense Stack

An AI agent is far more than a simple language model; it is a complex architecture comprising models, harnesses, guardrails, and logs. The alliance is focused on securing the entire "agent stack." Several members have already begun contributing critical technologies to this mission:

Key Technical Contributions

• NVIDIA: The company has introduced the NVIDIA Labs Object-Oriented Agent (NOOA), an open-source research framework available on GitHub. NOOA is designed to make agent behavior easier to test, trace, audit, and govern by integrating advanced safety capabilities directly into agent harnesses.
• HPE: Contributing to the SPIFFE/SPIRE project, HPE is helping to establish zero-trust identity standards that cryptographically verify AI agents, ensuring only authorized workloads can access enterprise resources.
• Hugging Face: The platform has offered Safetensors—a secure format for storing model weights—to the PyTorch Foundation, providing a transparent, tamper-proof method to prevent remote code execution.
• Microsoft: The introduction of MDASH (Multi-model Agentic Scanning Harness) allows for the orchestration of specialized agents to discover, debate, and prove the existence of exploitable bugs.
• SpaceXAI: By open-sourcing the Grok Build terminal-based coding agent and committing to release weights for the Grok model line, SpaceXAI is fostering greater trust and transparency within the developer community.

A Call to Policymakers

As governments and regulatory bodies around the world begin to draft AI safety frameworks, the alliance urges them to view open models and security tooling as defensive assets rather than liabilities. Blanket restrictions on open-source AI systems risk concentrating power and creating new, systemic vulnerabilities.

Instead, the alliance advocates for a policy shift: governments should treat shared open infrastructure—such as datasets, evaluation frameworks, and red-teaming tools—with the same strategic importance as the open-source software that currently powers the global internet.

The Future of AI Resilience

The era of AI agents holds the potential for unprecedented resilience, but only if we choose to build it on a foundation of shared security. The Open Secure AI Alliance believes that the future of technology will not be secured by secrecy, but by building systems that are robust enough to withstand public scrutiny and flexible enough to be improved by the collective intelligence of the global community.

By mobilizing the full spectrum of defenders, the alliance aims to ensure that the safety and security of artificial intelligence are not just features of a few closed systems, but a universal standard. The work of defending the AI era has begun, and the alliance invites researchers, governments, and industry leaders to join this critical mission.