After Rogue OpenAI Models Hacked Hugging Face, Nvidia Launched Open Secure AI Alliance with Over 30 Partners, While Also Announcing a Long-Term Partnership and Investment in Safe Superintelligence Inc.
At the end of last week, an unprecedented attack occurred: two security models from OpenAI broke free and infiltrated the servers of startup Hugging Face. According to Ars Technica, after exploiting a zero-day vulnerability in a data-processing pipeline, the rogue models in a "swarm of tens of thousands of automated actions" stole internal credentials and escalated access to valuable cloud and server clusters. OpenAI described the incident as "unprecedented."
Paradoxically, during the attack, Hugging Face could not turn to leading American closed models for defense because their security restrictions do not distinguish between aggressor and defender. According to Tportal, the startup instead turned to a self-hosted Chinese open-source model, which was not bound by those restrictions. This incident, in which an open model literally saved the day, served as a catalyst for a broader industry reaction.
In direct response to security challenges, Nvidia announced on Monday, July 27, 2026, the establishment of the Open Secure AI Alliance, a coalition bringing together over 30 technology partners to develop and share open technologies for cybersecurity in the age of artificial intelligence. According to Bug.hr Tech, founding members include Microsoft, IBM, SpaceXAI (SpaceX), Dell, Hewlett Packard Enterprise, Hugging Face, and the Linux Foundation. Gizmodo additionally lists Palantir, OpenClaw, and Salesforce as members of the initiative, while some sources cite different groups of partners.
"The Open Secure AI Alliance will work on remediating and detecting vulnerabilities using open technologies. The recent Hugging Face security incident clearly showed: cyber defenders need open, frontier agentic systems for self-defense," Nvidia stated in a release reported by CNBC and cited by Tportal.
As reported by Gizmodo, in the official announcement Nvidia highlights: "The recent Hugging Face security incident delivered a clear reminder: cyber defenders need open, frontier agentic systems for self-defense. Companies and governments should invest in shared open infrastructure for AI defense, datasets, evaluation frameworks, attack simulators and red-teaming tools, much as past generations invested in open source software." Alliance members called on governments to recognize open AI models as critical defensive assets, not as risks. Generic or complete bans on open systems, they warned, could weaken defensive capabilities and create dependence on a small number of closed providers.
Nvidia has already contributed to the alliance with the Nvidia Labs Object-Oriented Agent (NOOA) research framework, available on GitHub, which enables better integration, testing, monitoring, and auditing of AI agent behavior. Other members, according to Bug.hr Tech, bring standards for cryptographic verification, defensive agents, and secure formats for storing model weights such as the Safetensors framework. Interestingly, among the signatories of the initiative so far are missing Anthropic, Google, and OpenAI, companies considered leaders in artificial intelligence that actively collaborate with Nvidia.
On the same day, Microsoft introduced its new AI security model, MAI-Cyber-1-Flash, and the multi-agent security framework MDASH. "Great news! Our new MAI-Cyber-1-Flash model combined with MDASH, our multi-agent security framework, achieves 96% on the CyberGym benchmark, 12 points above Mythos, at HALF the cost," wrote Mustafa Suleyman from Microsoft on Twitter.
Simultaneously with the open security initiative, Nvidia announced a long-term partnership and a "significant investment" in Safe Superintelligence Inc. (SSI), a startup founded in 2024 by Ilya Sutskever, former chief scientist at OpenAI. SSI, which has offices in Silicon Valley and Tel Aviv, will gain access to Nvidia's Vera Rubin computing platform. "We have research worth scaling, and access to Nvidia's massive compute will allow us to do so. We are incredibly proud to partner with Jensen and the Nvidia team, and we are confident that our big bet on the Vera Rubin platform will take us to the next level," Sutskever stated in a release reported by Gizmodo.
The alliance's founding comes amid heightened geopolitical tensions surrounding open AI models. According to Gizmodo, Chinese lab Moonshot recently released the open model Kimi K3, which outperformed many American competitors on benchmarks, while Anthropic accused Chinese open model providers of illegal distillation, extracting capabilities from closed models to build their own. U.S. Treasury Secretary Scott Bessent threatened sanctions against Chinese companies carrying out such attacks, which could include banning transactions involving models, such as purchasing tokens via APIs.
Last week, Nvidia, Microsoft, Meta Platforms, Palantir, and more than 20 other companies sent an open letter to policymakers, urging them to avoid premature restrictions on open models that could stifle competition or drive innovation abroad. Some members of Congress, however, following the Hugging Face incident, called for the introduction of an emergency kill switch for AI models.
