'Open Secure AI Alliance' Formed for AI Safety and Security

Open source software is a core pillar of the global economy, spanning cloud, finance, manufacturing, and more, and it is a major beneficiary area especially in cybersecurity. In this context, the Open Secure AI Alliance was launched with the goal of leveraging open technology to identify and address vulnerabilities.

When it comes to AI security, the United States and its partners stand at a critical crossroads: whether to rely on a handful of closed systems for security infrastructure, or to build a defense system based on Open models and open tools that anyone can research and deploy.

Open models and open harnesses are essential to cybersecurity for the following reasons:

• Democratizing defense capabilities and increasing transparency
• Enabling cyber defense while protecting data at the same time
• Complementing closed frontier models through customization and local control capabilities
• Building a distributed, community-based defense system with no single point of failure

A recent Hugging Face security incident demonstrated the importance of open systems. When closed AI tools failed to distinguish between attackers and defenders, blocking forensic analysis, Hugging Face was able to run the GLM 5.2 open-weight model on its own infrastructure to analyze over 17,000 actions and block the intrusion.

The alliance's initial partners include global leaders in cloud, cybersecurity, and AI research such as NVIDIA, Adobe, Microsoft, NAVER, SK Telecom, CrowdStrike, and Red Hat. They plan to develop and share open technologies and tools to protect software and agents in the AI era.