NVIDIA launched the Open Secure AI Alliance on Monday with 37 partners, including Microsoft, IBM and Palantir, with the latest announcement confirming Korean companies Naver and SK Telecom as participating partners. According to reports from CNBC, the group has 37 companies in total, with Microsoft, IBM, Palantir, CrowdStrike, Red Hat, SpaceXAI and Hugging Face all joining. The alliance was formed in response to a July hack in which closed AI models refused to help Hugging Face investigate its own attacker. As reported by Business Standard, US companies leading the AI race, including OpenAI, Anthropic and Google, are not part of the founding alliance, highlighting the strategic focus on open-source security solutions. The coalition brings together Nvidia, Microsoft, SpaceX, Palantir and a broad roster of American and European technology firms, with its stated purpose being to build and share open-source tools that strengthen the security of open AI systems. Members will share open AI models, data and security tools, with NVIDIA stating that defenders need AI they can open up, change and run themselves.
The attack on Hugging Face began with a poisoned dataset that let an attacker run code on one of the company's machines. According to Hugging Face's disclosure reported by CNBC, the attacker then stole passwords and spread through internal systems over a weekend. Hugging Face's own AI spotted the break-in, but when the company asked leading commercial AI models to study the attack, they refused. The company used an open model called GLM 5.2 instead, which sorted more than 17,000 attacker actions in hours, not days, with no passwords left the building. On July 16, Hugging Face said it did not know which AI ran the attack, but five days later, OpenAI supplied the answer that two of its models did it during a security test with safety limits switched off. The OpenAI-Hugging Face incident showed a practical truth: when defenders cannot inspect, adapt and run advanced AI on their own infrastructure, their ability to respond is constrained at exactly the moment speed matters most. As reported by Business Standard, Hugging Face instead turned to a self-hosted, open-weight model developed in China, which was not constrained by the same restrictions and enabled it to repel the attack. The incident, which OpenAI described as "unprecedented," has raised concerns over whether enterprise systems are prepared for increasingly autonomous AI agents that can identify vulnerabilities, chain exploits and launch cyberattacks with little or no human intervention.
Major global tech companies and research institutions, including NVIDIA, Microsoft, Adobe, Cisco, CrowdStrike, Databricks, HPE, IBM, Palantir, Palo Alto Networks, Red Hat, Salesforce, and SpaceX AI, have joined this consortium as founding members. Among Korean companies, Naver and SK Telecom are participating as partners and plan to collaborate on the development of agent protection and defense technologies. The alliance's participating companies plan to build an open defense stack that includes AI agent identity, isolation, secure model formats, multi-model scanning, and secure coding workflows. NVIDIA has released "NVIDIA Labs Object-Oriented Agent (NOOA)", a research framework designed to assist in testing, tracking, auditing, and managing agent behavior, on GitHub. Based on the Linux Foundation's Akrites initiative and OpenSSF community activities, the alliance plans to work together to address and disclose vulnerabilities using open technologies. As reported by Business Standard, "The world needs both closed and open models. For cybersecurity, open models and open harnesses are essential because they democratize defensive capabilities, increase transparency for defenders, enable cyber defense while protecting data, and complement frontier closed models with customizable, localized controls." The alliance emphasizes that "open source enables massively distributed community-driven and self-controlled defense—with no single point of failure."
The alliance now includes 37 participating companies with major additions including Capital One, Cloudflare, Cloudera, DoorDash, Elastic, NetApp, ServiceNow, SAP, Snowflake, Synopsys, Thinking Machines Lab and TrendAI. NVIDIA will provide the Open Secure AI Alliance with open models and model weights, data and new agent harness research to speed up the development of new cybersecurity tools and technologies. The NOOA research framework helps the harness integrate effectively with models, making it easier to test, track, audit and manage agent behavior. Alliance participants aim to build an open agent defense stack spanning identity and isolation through secure model formats, multimodel scanning and secure coding workflows. Members are contributing specific tools, with Microsoft contributing MDASH, a system that runs multiple AI agents to find exploitable bugs, while SpaceXAI open-sourced its Grok Build coding agent and plans to release the weights of its Grok models. As reported by Business Standard, "Companies and countries need open frontier defensive tools and techniques so critical industries can build security systems across a multi-vendor ecosystem and avoid single points of failure." The mission of the Open Secure AI Alliance is "to ensure defenders everywhere have open, frontier tools they can trust and control."
The alliance carries a policy message regarding potential restrictions on open AI. According to CNBC, NVIDIA says a broad ban on open AI would leave defenders weaker and hand control to a few big closed companies. The launch comes as US lawmakers consider measures to restrict the growing use of Chinese AI systems, with Treasury Secretary Scott Bessent stating that Chinese firms found to be conducting distillation attacks against American companies could face sanctions. As reported by Business Standard, NVIDIA stated, "We must not assume that AI safety can be ensured through secrecy alone," adding, "We hope that governments, industry, and researchers around the world will work together to safeguard the AI era by building systems that are robust enough to withstand scrutiny and open enough to unite the entire community of defenders." The complication for policymakers is that many of the most advanced open-weight models currently available are produced by Chinese developers, raising fears that a heavy-handed approach could inadvertently damage the wider open-source ecosystem. NVIDIA emphasized, "We hope that governments, industry, and researchers around the world will work together to safeguard the AI era by building systems that are robust enough to withstand scrutiny and open enough to unite the entire community of defenders."
