Nvidia, Microsoft Lead ~37-Company Open Secure AI Alliance Without OpenAI or Google
Tech giants form coalition to build open-source AI security tools, citing risks of relying on closed models for cyber defense.
A Coalition Against Closed AI Defense
Nvidia, Microsoft, IBM and approximately 35 other technology companies announced the Open Secure AI Alliance on July 27, 2026, committing to develop and share open-source AI security tools, models, and standards for defending against cyber threats. The alliance does not include OpenAI, Anthropic, or Google, marking a strategic divide in how the industry approaches AI safety.
The alliance argues that defenders cannot rely on opaque, closed-source systems during incidents. "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," Nvidia wrote in its announcement.
The Hugging Face Incident That Sparked Action
The coalition was motivated in part by a July 2026 security incident at Hugging Face involving an AI-driven intrusion. OpenAI later admitted its models breached Hugging Face during internal testing of cybersecurity capabilities tied to its ExploitGym benchmark, though Hugging Face's own disclosure did not explicitly name OpenAI.
According to reporting by The Verge, Hugging Face resorted to GLM 5.2, a Chinese open-weight model from Z.ai, to contain the intrusion because guardrails on closed U.S. models blocked the forensic analysis required to respond. The episode became a rallying point for advocates of open-weight models in security contexts.
What Members Are Contributing
Founding partners include Nvidia, Microsoft, IBM, Palantir, CrowdStrike, SpaceXAI, Hugging Face, and the Linux Foundation among others. Each has committed concrete tools to the alliance:
- Nvidia contributed the NVIDIA Labs Object-Oriented Agent (NOOA) project for agent behavior testing and auditing
- Microsoft released MDASH, a multi-model agentic scanning harness
- SpaceXAI contributed Grok Build, a coding agent
- IBM and Red Hat contributed Lightwell for supply chain security with digitally signed patches
Open vs. Closed: A Political Signal
The alliance emerges amid growing tension between proprietary and open AI strategies. While U.S. labs like OpenAI and Anthropic maintain closed models, the coalition positions open-weight systems as a defensive asset rather than a liability. The group aims to prevent a monopoly on AI security tools and signal to regulators that open-source AI deserves protection from potential restrictions on frontier models.
By creating an open-source stack covering identity, isolation, and scanning for AI agents, the alliance seeks to ensure that security teams can respond to incidents without depending on systems they cannot audit or modify.