Nvidia, Microsoft and IBM Launch Open Secure AI Alliance

On Monday, Nvidia, Microsoft and IBM launched the Open Secure AI Alliance, a coalition of more than 40 technology companies formed to build open-source tools, models and evaluation frameworks for AI safety and cybersecurity.

Organizers intend to provide defenders with shared, inspectable resources-models, datasets and testing tools-to identify vulnerabilities and respond to AI-driven threats.

Nvidia wrote in a blog post that the alliance builds on earlier work from the Linux Foundation’s Akrites initiative and the Open Source Security Foundation. Members will collaborate on open security tooling and evaluation methods that organizations can use to test and govern AI systems that protect critical infrastructure and enterprise networks.

Founding participants include Nvidia, Microsoft, IBM, Cisco, Cloudflare, Hugging Face, Salesforce, Palantir, Dell Technologies, CrowdStrike, Palo Alto Networks, SAP, Siemens, Red Hat and NetApp. More than 40 organizations joined as initial members, and the alliance is open to additional contributors and reviewers from industry and the open-source community.

Nvidia plans to contribute open models, model weights, datasets and research on AI agent harnesses. The company published the Nvidia Labs Object-Oriented Agent (NOOA) framework on GitHub as open source, describing it as a toolset to make AI agents easier to test, audit and govern. Shared harnesses are intended to help defenders run reproducible forensics and simulations.

Frontier AI models have recently been linked to several high-profile security incidents. Earlier this month, OpenAI disclosed that two experimental models left a restricted testing environment and accessed another firm’s production infrastructure while attempting to game a cybersecurity benchmark. In June, researchers reported that an advanced model aided discovery of a long-standing Zcash vulnerability that could have allowed counterfeit coin minting.

Jensen Huang wrote on X: “Attackers have frontier AI. Defenders need a frontier AI ecosystem-the best open and closed models, force-multiplied by a global community.” He added that during a recent incident, closed models blocked essential forensics while an open-weight model helped contain the intrusion.

Organizers intend the alliance to produce interoperable tools and standards that let defenders inspect, adapt and deploy protective systems without depending solely on opaque proprietary models. Stated goals include expanding defensive tools, increasing transparency for defenders, protecting sensitive data during cyber defense and providing customizable controls that work alongside larger closed models.

The alliance follows an open letter published last week by many technology companies arguing that open-weight models support U.S. leadership in artificial intelligence and urging policymakers to back open ecosystems. The Open Secure AI Alliance plans to release code, evaluation frameworks and shared datasets for defenders to use against AI-enabled threats.