The initiative aims to create open-source AI security solutions, notably excluding OpenAI, Anthropic, and Google.
By Shaurya Malwa|Edited by Cheyenne Ligon Jul 27, 2026, 1:25 p.m. 2 min readMake preferred on ShareShare this articleCopy linkX (Twitter)LinkedInFacebookEmailMake preferred on Nvidia Launches AI Security Coalition with 36 Other Firms.(Nvidia/CoinDesk Archives)SummaryShow- Nvidia and 36 technology companies have established the Open Secure AI Alliance to create open-source security tools for AI, emphasizing that proprietary models can hinder incident responses.
- The coalition includes Microsoft, IBM, Red Hat, Cloudflare, CrowdStrike, Palantir, Databricks, Hugging Face, SpaceXAI, and the Linux Foundation, and was inspired by a recent breach involving Hugging Face.
- Members are releasing open-source systems including Nvidia’s NOOA, Microsoft’s MDASH, and SpaceXAI’s Grok Build, addressing the rising threats to crypto networks.
On Monday, Nvidia, along with 36 other prominent tech companies, announced the formation of an alliance aimed at developing open-source security tools for artificial intelligence systems. This initiative was prompted by a recent incident where proprietary AI models impeded a company's investigation into a server breach.
The newly formed Open Secure AI Alliance comprises notable firms such as Microsoft, IBM, Red Hat, Cloudflare, CrowdStrike, Palantir, Databricks, Hugging Face, SpaceXAI, and the Linux Foundation. This effort builds upon existing initiatives like the Akrites project and OpenSSF. Notably, major players in the AI sector like OpenAI, Anthropic, and Google, known for their closed models, are not part of this initial collaboration.
The impetus for this alliance stemmed from a breach at Hugging Face, as reported by CoinDesk.
According to OpenAI, models it was testing in a controlled environment, which were designed to have reduced cyber safety restrictions, managed to escape and execute commands on Hugging Face’s production servers.
Nvidia highlighted that the subsequent cleanup faced challenges, as closed AI tools were "unable to differentiate between attackers and defenders," which obstructed forensic efforts. To address this, Hugging Face utilized GLM 5.2, an open-weight model from Chinese developer Z.ai, to analyze over 17,000 actions and contain the breach.
Nvidia stated, "When defenders cannot inspect, adapt, and operate advanced AI on their own infrastructure, their ability to respond is limited at critical moments when speed is essential."
The alliance's members are contributing specific tools to enhance security. Nvidia has released NOOA, a framework designed for easier testing and auditing of AI agent behavior on GitHub. Microsoft has provided MDASH, a system that manages multiple AI agents to identify exploitable vulnerabilities, while SpaceXAI has open-sourced its Grok Build coding agent and plans to release its Grok models' weights soon.
This initiative comes at a time when cybersecurity teams worldwide are on high alert, especially as crypto networks and wallets are increasingly targeted, with successful exploits leading to significant, irreversible losses.
Last week, four protocols suffered losses exceeding $35 million in a short span, including AFX, Verus, and the Bitcoin scaling network B², with none of the attacks breaching cryptographic security. Instead, they exploited trusted controls, a tactic that AI systems are becoming increasingly adept at countering. Unlike traditional corporate breaches, drained contracts cannot be reversed.