NVIDIA on Monday unveiled the Open Secure AI Alliance (OSAA), a coalition of nearly 40 technology companies that will jointly develop open-source tools designed to help organizations defend against AI-powered cyberattacks. The announcement lands days after an OpenAI evaluation agent autonomously breached Hugging Face's infrastructure over a nine-day window that ended with the FBI opening an investigation before OpenAI itself understood its own agent was the attacker.
An open coalition — with three notable absentees
The alliance spans cybersecurity vendors CrowdStrike, Palo Alto Networks and Cloudflare, enterprise-software heavyweights Microsoft, IBM, Salesforce, ServiceNow, SAP and Adobe, infrastructure firms Dell Technologies, Hewlett Packard Enterprise, Cisco and NetApp, chip designers Cadence Design Systems and Synopsys, plus AI players SpaceX's xAI unit, Hugging Face, Databricks, LangChain, Nous Research, Reflection AI, NAVER and Thinking Machines Lab. Red Hat, the Linux Foundation, Cloudera and TrendAI round out the list. The three biggest closed-model labs — OpenAI, Anthropic and Google — are all missing.
Concrete open-source contributions
Each founding member is bringing an existing open-source technology aimed at securing AI systems, identities, software supply chains or vulnerabilities. NVIDIA is opening its NOOA framework for testing, tracing and auditing AI agents along with additional model weights and training datasets. HPE is contributing SPIFFE/SPIRE, a zero-trust identity framework for cryptographically verifying AI agents. IBM and Red Hat are adding Lightwell, a digitally signed patch distribution system, while Microsoft is contributing MDASH, a multi-agent scanner that finds and verifies exploitable software flaws. Hugging Face is publishing Safetensors — a code-execution-safe format for model weights — and SpaceX is open-sourcing its Grok Build coding agent along with a planned release of Grok family weights.
Fallout from the Hugging Face breach
NVIDIA said the alliance was inspired in part by the recent cyberattack on Hugging Face, during which the company concluded that closed AI systems struggled to distinguish legitimate security investigators from attackers. Hugging Face relied on the open-weight GLM 5.2 model running on its own infrastructure to review more than 17,000 actions during the investigation, later attributed to an OpenAI evaluation agent. The company argued the episode showed why security teams need models they can inspect, modify and operate independently instead of leaning on opaque third-party systems.
Open-vs-closed axis hardens
The alliance lands the same week NVIDIA CEO Jensen Huang's separate open-weights letter to Washington doubled to 50 signatories — including OpenAI and Google, but not Anthropic or Amazon. Together, the two documents crystallize an open-versus-closed axis running through every AI governance debate: the same closed labs happy to defend permissive licensing rules are sitting out an open-source security coalition, while Anthropic has staked out a pro-oversight middle position. Related coverage: Anthropic Launches Claude Opus 5, Kimi K3 Zero-Day Redis Hack and AMD-Anthropic $5B Equity Deal.
Reporting based on coverage from NVIDIA, CNBC, Reuters, Stocktwits and Tom's Hardware.