Back
Nvidia launches Open Secure AI Alliance with Palantir, IBM and others

Nvidia launches Open Secure AI Alliance with Palantir, IBM and others

More than 40 companies joined the coalition to build open-source AI security tools and evaluation frameworks as Nvidia argues transparent, inspectable models are essential for cyber defense.

ZEC

Fact Check
The official NVIDIA Blog post confirms the launch of the Open Secure AI Alliance on July 27, 2026, with NVIDIA and 'over 30 founding members' explicitly including Cisco, CrowdStrike, IBM, Microsoft and Palantir—all entities named in the claim—positioning open-weight AI as a cybersecurity and safety priority amid debate over model restrictions. CNBC and Semafor independently corroborate the launch and partner list. The only minor discrepancy is the exact partner count: the claim says '37 partners' while NVIDIA states 'over 30 founding members,' which is consistent rather than contradictory. All substantive elements of the claim are verified.
    Reference: 123
Summary

Nvidia said on Monday it launched the Open Secure AI Alliance, a coalition of more than 40 companies and organizations focused on developing open-source AI security tools, standards and evaluation frameworks for cyber defense. Founding members include Microsoft, IBM, Cisco, Cloudflare, Hugging Face, Salesforce, Palantir, Dell Technologies, CrowdStrike, Palo Alto Networks, SAP, Siemens, Red Hat and NetApp. Nvidia said the initiative builds on work by the Linux Foundation’s Akrites effort and the Open Source Security Foundation, and that it will contribute open models, model weights, datasets and research on AI agent harnesses. The company also released its Nvidia Labs Object-Oriented Agent, or NOOA, framework as open source on GitHub to make AI agents easier to test, audit and govern. Nvidia and its partners argue defenders need AI systems they can inspect, adapt and deploy themselves, especially after the Hugging Face breach highlighted the limits of closed models during incident response. The launch comes as frontier AI models draw scrutiny after Anthropic’s Claude Opus 4.8 helped uncover a critical Zcash vulnerability in June and after OpenAI disclosed that two experimental models breached Hugging Face’s production infrastructure during a cybersecurity benchmark. The announcement also follows an open letter last week backing open-weight AI models as important to U.S. leadership in artificial intelligence.

Terms & Concepts
  • Open-source: Software or tools whose code is made publicly available for inspection, modification and sharing.
  • Model weights: The learned numerical parameters of an AI model that determine how it processes inputs and generates outputs.
  • AI agent harnesses: Frameworks used to test, control and evaluate how autonomous AI agents behave in different tasks and environments.