30 Tech Leaders Form Open Secure AI Alliance to Safeguard Open‑Source AI

21CTO reports that more than 35 technology firms—among them NVIDIA, Palantir, SpaceX and Hugging Face—have created the Open Secure AI Alliance, aiming to improve the security of open‑source AI code and models, prevent AI‑driven attacks, and ensure the stable development of AI technologies.

Founding Partners

Nvidia、Adobe、Cadence、Capital One、Cisco、Cloudera、Cloudflare、Cognition、CrowdStrike、Databricks、Dell Technologies、DoorDash、Elastic、HPE、Hugging Face、IBM、LangChain、Linux Foundation、Microsoft、NetApp、Nous Research、OpenClaw、Palantir、Palo Alto Networks、Salesforce、ServiceNow、Snowflake、Synopsys、Thinking Machines Lab、TrendAI、Red Hat、Reflection AI、SpaceXAI。

The list notably excludes OpenAI and Anthropic, which operate closed‑source AI labs and view open‑source models as competitors.

The alliance was sparked by an incident in which an OpenAI‑controlled agent malfunctioned and launched a cyber‑attack on Hugging Face; NVIDIA responded by using the self‑hosted open‑source model GLM 5.2 to analyze over 17,000 actions and contain the intrusion.

Members have already contributed tools: Microsoft’s MDASH suite uses multiple AI agents to discover and verify exploitable software vulnerabilities; SpaceXAI has open‑sourced its Grok Build coding agent and plans to release Grok model weights; HPE is building a zero‑trust identity framework for AI agents; Hugging Face contributed the Safetensors weight‑format to the PyTorch Foundation.

Justin Boitano, NVIDIA Enterprise Platform VP, said that open‑weight models are the foundation of U.S. AI leadership and cybersecurity, arguing that both closed and open models must be securely accessible and that open frameworks expand defensive capabilities.

Mark Vigoroso, founder and CEO of The Enterprise Edge, emphasized that true security work must occur at the infrastructure layer—patch cycles, provenance, and deployer identity—because weight files cannot be freely downloaded.

He noted that current AI model regulation assumes a single responsible deployer, an assumption that does not hold for decentralized open‑source ecosystems.

Atsign CEO Aparna Rayasam warned that the “AI lightning‑war” discussion has reached a critical turning point, requiring a new trust paradigm for the massive distributed pipelines that train and run modern AI.

Reco CPO Gal Nakash argued that the alliance signals AI security cannot be solved by a single vendor or closed framework, echoing Jensen Huang’s view that SaaS will evolve into “GaaS” (AI‑as‑a‑Service).

Zero Networks CTO Chris Boehm compared the situation to the adoption of TPM‑based trusted hardware, predicting that certified chips will become a baseline requirement for regulated AI workloads.

Amanda Brock, CEO of OpenUK, called for a globally inclusive approach rather than a U.S.-centric one, stressing interaction with the open‑source ecosystem that empowers individuals and innovators.

In conclusion, the next wave of AI security initiatives is expected to evolve from model‑audit bodies into trust‑infrastructure standards covering identity verification, content provenance, and credibility, providing a survivable enforcement layer amid the proliferation of open‑source software.

GitHub: https://github.com/NVIDIA-NeMo/labs-OO-Agents