The Open Secure AI Alliance is an industry coalition launched by NVIDIA on July 27, 2026 with more than 25 inaugural partners. Its stated position is that regulators should treat open models as "defensive assets, not liabilities" (Source: blogs.nvidia.com). It was announced three days after the "Open Weights and American AI Leadership" letter (Open Weights and American AI Leadership (industry letter, July 2026)), converting that letter's ad hoc signatory roster into a standing organization.
Membership
Inaugural partners named at launch are Adobe, Cisco, CrowdStrike, Databricks, GitHub, Hugging Face, IBM, The Linux Foundation, Microsoft, Mistral, Palantir, Palo Alto Networks, Perplexity, Red Hat, Salesforce, SAP, SpacexAI, Thinking Machines Lab, and Zscaler (Source: blogs.nvidia.com; wsj.com). The security-vendor share of the roster is larger than in the July 24 letter, consistent with the alliance's framing of open weights as a defensive-capability question.
Anthropic is not a member and did not sign the preceding letter (Source: axios.com).
Positions and activities
The alliance's launch argument draws directly on the July 2026 Hugging Face security incident. NVIDIA stated that closed commercial models refused the forensic work required after the intrusion, so Hugging Face ran the open-weight GLM 5.2 on its own infrastructure to analyze more than 17,000 recorded actions (Source: blogs.nvidia.com; huggingface.co). The same episode had earlier been used by Ben Thompson to argue for loosening cybersecurity restrictions on US closed models (Who's Afraid of Chinese Models? (Ben Thompson, Stratechery, July 2026)).
NVIDIA said it is contributing open models, weights, data and agent-harness research, including the NVIDIA Labs Object-Oriented Agent project released on GitHub. The alliance's stated aim is to have the U.S. government and its allies treat open-source AI as a cybersecurity asset rather than a threat (Source: blogs.nvidia.com; wsj.com).
The alliance's premise runs against the UK AI Security Institute position that open release "precludes many of the safety measures that closed model developers can use to detect and disrupt misuse" (How Far Behind the Frontier are Leading Open Weight Models on Cyber? (UK AISI, July 2026)). Both positions rest on the same observation — that open weights permit unrestricted local use — and differ on whether defenders or attackers gain more from it.
Relationships
- supports: Open Weights and American AI Leadership (industry letter, July 2026) — same policy position, institutionalized.
- contradicts: How Far Behind the Frontier are Leading Open Weight Models on Cyber? (UK AISI, July 2026) — on whether open release is net-defensive.
- related: Open-Weight Frontier Models, Dual-Use Frontier AI, AI and Cybersecurity.
- related: Nvidia & TSMC — AI Compute Infrastructure, Hugging Face, Microsoft.