Nvidia, Microsoft, and IBM have teamed up with more than 40 other technology companies and organizations to form the Open Secure AI Alliance. The group aims to develop open-source AI security tools and standards for cyber defense. The announcement comes as businesses and governments race to protect AI systems from a growing wave of attacks.
Why the alliance was formed
The alliance wants to make it easier for organizations to build and deploy secure AI. Right now, many companies rely on proprietary security tools that don't always work well together. By creating open-source standards, the group hopes to close gaps that attackers can exploit. The focus is on practical defenses — things like detecting poisoned training data, spotting adversarial inputs, and securing the supply chain for AI models.
Who's in the room
Besides the three founding companies, the alliance includes a wide range of players: cloud providers, chipmakers, cybersecurity firms, and academic institutions. Names like Amazon Web Services, Google Cloud, Meta, and OpenAI are not on the list, but the group says it's open to new members. The diversity of the current members — from Arm and Cisco to the Linux Foundation and the University of California, Berkeley — suggests the effort is meant to be broad rather than exclusive.
The open-source approach
The alliance plans to release reference architectures, toolkits, and best practices under open-source licenses. That means any organization can inspect, modify, and use the code without paying licensing fees. The hope is that this transparency will lead to faster innovation and wider adoption of security measures. It also means the community can audit the tools for vulnerabilities, which is a double-edged sword: openness helps find bugs, but it also gives attackers a clear view of the defenses.
The alliance hasn't announced a specific product or release date yet. Members are expected to start contributing code and documentation in the coming months. The first projects will likely focus on securing the machine learning pipeline — from data collection to model deployment. A key question is how the group will handle governance and updates once the tools are out in the wild. The alliance says it will operate under a neutral foundation, but the details are still being worked out.
