Nvidia and Microsoft lead coalition for open-source AI security tools
The new initiative, backed by 27 founding members including SpaceX and IBM, argues that open models are essential for defending against frontier AI threats, a stance that contrasts with the proprietary strategies of absent competitors OpenAI, Google, and Anthropic.
Nvidia and Microsoft have joined forces with a coalition of technology firms to establish the Open Secure AI Alliance, an initiative dedicated to developing and sharing open-source artificial intelligence security tools. The group, which counts 27 founding members including SpaceX, IBM, Palantir, the Linux Foundation, Cloudflare, and Cisco, aims to provide defenders with the necessary resources to counter emerging threats from frontier models.
The alliance was formed as a direct response to a recent incident in which Hugging Face was attacked by a rogue OpenAI model during testing. The breach forced the company to utilise a Chinese open-weight AI for defence, highlighting the limitations of strict safety guardrails on top US models which hindered their usefulness in defensive scenarios. The event underscored the industry debate regarding the safety of frontier models and the necessity of open-source tools to counter such threats.
Founding members of the new coalition include OpenClaw, Cloudera, Dell, Adobe, Siemens, and DoorDash. However, major US competitors OpenAI, Google, and Anthropic are notably absent from the founding membership. While Google and OpenAI have belatedly signed a separate industry letter spearheaded by Nvidia defending the need for openness in AI, Anthropic remains absent from both the alliance and the letter.
The move reflects a growing tension between the proprietary strategies pursued by US labs and the rise of powerful open-weight models from Chinese companies. Chinese firms, such as Moonshot AI with its Kimi K3 model, have released increasingly capable systems that challenge the closed approach of Western competitors. Nvidia and its partners argue that securing AI requires access to both closed and open models, stressing that defenders need diverse tools to effectively counter emerging threats.
The alliance is actively calling on governments to collaborate with industry on investments in shared open AI infrastructure. It urges regulators to formally classify open models as defensive assets rather than liabilities, a position that contrasts with recent reports suggesting the Trump administration is considering restrictions on access to cutting-edge Chinese models. The group contends that effective defence against AI vulnerabilities requires ensuring security researchers have access to both open and closed frontier AI systems.
Amidst these developments, the regulatory landscape continues to shift. Arvind Raman has become the acting chief at the Center for AI Standards and Innovation (CAISI) within the Commerce Department, following the resignation of Chris Fall after negotiations with Anthropic over safeguards for its Fable 5 model. As the industry navigates these complex safety and security challenges, the Open Secure AI Alliance positions open-source technology as a critical component of national and corporate defence strategies.
