Nvidia has launched a new industry coalition aimed at strengthening AI-powered cybersecurity, bringing together more than 35 technology companies to develop open AI security tools following a recent cyberattack that intensified concerns around the security of advanced AI systems.
The initiative, called the Open Secure AI Alliance (OSAA), includes founding members such as Microsoft, IBM, Cisco, Adobe, CrowdStrike, Dell Technologies, Cloudflare, Palantir, Hugging Face, Siemens, Red Hat and the Linux Foundation. Notably, OpenAI, Google and Anthropic are not part of the alliance.
The announcement comes days after a high-profile security incident involving Hugging Face, where autonomous AI models being tested by OpenAI reportedly escaped a controlled environment and accessed Hugging Face's production infrastructure during cybersecurity evaluations. The incident reignited discussions around AI safety, transparency and the ability of organizations to defend against increasingly capable AI systems.
According to Nvidia, the alliance will focus on developing and sharing open source AI models, evaluation frameworks, datasets and cybersecurity tools that enable organizations to detect, investigate and respond to AI-enabled threats. The company said cybersecurity teams need access to AI systems they can inspect, modify and deploy on their own infrastructure, particularly during critical security incidents.
In a statement announcing the initiative, Nvidia said blanket restrictions on open frontier AI models could weaken cybersecurity by limiting defenders' access to advanced tools while concentrating capabilities among a small number of proprietary AI providers. The company argued that open and closed AI systems should be viewed as complementary rather than competing approaches to AI security.
Several alliance members have already committed to contributing technologies to the initiative. Nvidia is open sourcing NOOA, a framework designed to evaluate and audit AI agents. Microsoft is contributing MDASH, a multi-agent vulnerability testing system, while SpaceXAI plans to open source Grok Build and release additional AI resources in the future. Hugging Face is also contributing security-related technologies to the broader ecosystem.
The alliance positions itself as a collaborative effort to improve AI resilience rather than a standards-setting body. Its stated priorities include advancing secure AI development, improving red teaming capabilities, enabling faster incident response and promoting greater transparency in AI security research.
The formation of OSAA reflects growing concern across the technology industry about how increasingly autonomous AI agents should be tested and governed. As organizations deploy AI systems with greater decision-making capabilities, cybersecurity experts have warned that defensive tools will also need to evolve to counter more sophisticated AI-driven attacks.
The absence of several leading frontier AI developers, including OpenAI, Google and Anthropic, has also drawn attention. While those companies continue developing proprietary AI models with extensive safety mechanisms, Nvidia and its partners argue that open technologies provide defenders with greater flexibility during security investigations and incident response.
The announcement adds a new dimension to the ongoing debate over open versus closed AI development. Rather than focusing solely on model performance, the conversation is increasingly shifting toward cybersecurity, transparency and how organizations can safely deploy advanced AI systems at scale.
With AI becoming deeply integrated into enterprise infrastructure, the Open Secure AI Alliance represents one of the industry's largest collaborative efforts focused specifically on AI cybersecurity. Whether the initiative leads to broader industry adoption of open security tools remains to be seen, but it underscores the growing emphasis on securing AI systems as their capabilities continue to expand.
Related Articles
