Nvidia and more than 30 technology companies are joining forces to build open-source artificial intelligence tools for cyber defence, stepping directly into a widening dispute over whether freely available AI models make the digital world safer or more dangerous.
The new Open Secure AI Alliance – backed at launch by Microsoft, IBM, Palantir, CrowdStrike, Cisco, Dell and the Hugging Face model-sharing platform, among others – plans to create and distribute tools that security teams can examine, adapt and operate inside their own systems, the group said in a statement.
The push follows a breach this month at Hugging Face that OpenAI has described as the first publicly disclosed instance of an AI model autonomously carrying out a real-world cyberattack.
OpenAI called the episode “unprecedented,” saying two of its models broke out of a sandboxed testing environment during an internal evaluation, reached the open internet and compromised Hugging Face’s infrastructure.
To stop the intrusion, the company turned not to its own systems but to GLM 5.2, an open model from the Chinese firm Zhipu AI, also known as Z.ai.
The fact that a Chinese model succeeded where American models hesitated has added fresh urgency to a debate already simmering in Silicon Valley and Washington, where policymakers are considering limits on the use of Chinese AI systems as the technology rivalry with Beijing deepens.
An AI agent went rogue during testing – should we be worried?
Much of Silicon Valley has resisted those proposed limits, with many industry voices arguing that restrictions on open-source AI would unfairly protect OpenAI and Anthropic, the dominant companies selling the world’s most powerful closed models.
Anthropic, the company behind the Claude chatbot, has faced particular scrutiny as the most prominent holdout from publicly backing the open-source cause since the Hugging Face breach.
In announcing the alliance, its members pressed regulators to view open models as “defensive assets, not liabilities,” cautioning that sweeping restrictions could leave cyber defenders weaker while pushing power into the hands of a small group of closed AI providers.
The companies said they would contribute tools aimed at strengthening collective cyber defence.
Nvidia said it plans to release open models, data and new research on AI agent software, while Microsoft said it would provide technology designed to help AI agents identify software bugs.
Elon Musk’s SpaceXAI said it had open-sourced its Grok coding agent and committed to publishing the source data behind its Grok models.
