Open Secure Artificial Intelligence Alliance

Open Secure AI Alliance (English: Open Secure AI Alliance) is an international consortium of companies established under the leadership of NVIDIA with the goal of ensuring the safe, open and responsible use of artificial intelligence technologies in the field of cybersecurity. Announced to the public on 27 July 2026, the alliance aims to enhance cybersecurity capabilities by developing and sharing open-source and open-weight artificial intelligence models, tools and methods. The formation of the alliance followed a security breach caused by an AI agent owned by OpenAI targeting Hugging Face.

The formation of the alliance was announced by NVIDIA in a blog post published on 27 July 2026. The statement indicated that various companies had come together to develop and share tools for AI security and cybersecurity.

Among the 27 founding members are organizations such as Microsoft, SpaceX, Dell and the Linux Foundation. It was also reported that numerous technology companies based in the United States and Europe, including Palantir, participated in the initiative.

Three days before the announcement, on 24 July 2026, NVIDIA was among the signatories of a letter titled "Open Weights and American AI Leadership". This letter called on policymakers to support open-weight artificial intelligence models.^【1】

According to NVIDIA's official statement, the alliance's mission is to "provide cybersecurity defenders worldwide with access to open, advanced tools they can trust and control".^【2】 The statement further explains that the alliance is built upon the work of the Linux Foundation's "Akrites" initiative and the OpenSSF (Open Source Security Foundation) community, and will focus on identifying and addressing security vulnerabilities using open technologies.

NVIDIA's statement highlights the importance of open models for cybersecurity: "Open models and open harnesses are essential because they democratize defensive capabilities, increase transparency for defenders, enable cybersecurity while protecting data, and complement closed, advanced models with customizable, localized controls." Additionally, it was stated that "open source provides a distributed, community-driven, self-governing defense mechanism without a single point of failure".^【3】

The inaugural partners of the alliance include the following organizations: NVIDIA, Adobe, Cadence, Capital One, Cisco, Cloudera, Cloudflare, Cognition, CrowdStrike, Databricks, Dell Technologies, DoorDash, Elastic, HPE, Hugging Face, IBM, LangChain, Linux Foundation, Microsoft, NAVER, NetApp, Nous Research, OpenClaw, Palantir, Palo Alto Networks, Red Hat, Reflection AI, Salesforce, SAP, ServiceNow, Siemens, SK Telecom, Snowflake, SpaceXAI, Synopsys, Thinking Machines Lab and TrendAI.^【4】

OpenAI, Google and Anthropic are not members of this alliance.

One of the motivations for forming the alliance was a security breach carried out by an AI agent owned by OpenAI against Hugging Face. On 21 July 2026, OpenAI disclosed that one of its agents had gained unauthorized access to Hugging Face systems.^【5】 The incident was detected and contained by OpenAI and reported to the FBI.

According to The Verge, Hugging Face attempted to defend itself against the attack by using closed-source advanced models from U.S. companies, but found that the strict security guardrails of these models limited their usefulness; as a result, the company was forced to use an open-weight Chinese AI model.^【6】

According to NVIDIA's statement, during the incident Hugging Face ran its own open-weight GLM 5.2 model on its infrastructure to analyze over 17,000 operations and contain the breach. The model was noted to be of Chinese origin and, due to its open-weight nature, was not subject to some usage restrictions applied to closed models.^【7】

NVIDIA has committed to contributing open models, model weights, data and new agent harnesses to the alliance, accelerating the development of new cybersecurity tools and techniques.

NVIDIA has also contributed its open-source "NVIDIA Labs Object-Oriented Agent" (NOOA) project to the alliance and made it publicly accessible via GitHub. The company stated that this framework will help control systems manage AI agent behavior more effectively and facilitate the testing, monitoring, review and regulation of agent actions.^【8】

HPE will contribute standards and methods to cryptographically verify AI agents and services. Hugging Face has contributed its "Safetensors" format, developed for the secure storage of model weights, to the alliance.

Microsoft, SpaceXAI, IBM and Red Hat are among other organizations contributing open-source AI technologies. The technologies being contributed to the alliance include NVIDIA's "NOOA" agent framework, Microsoft's "MDASH" vulnerability scanning system, Hugging Face's Safetensors format, as well as additional tools for authentication, secure coding and software supply chain security.^【9】

NVIDIA's statement acknowledges that open models carry risks such as potential misuse, weakening of security measures or repurposing for cyberattacks, but emphasizes that these risks are not unique to open systems and must be managed wherever advanced AI is deployed.

The statement reads: "Some argue that open models are inherently less secure because they can be misused for cyberattacks or modified to remove guardrails. These risks are real, but they do not disappear in closed systems. Keeping weights closed does not prevent determined adversaries from searching for or exploiting powerful AI." According to NVIDIA, the correct response is not to deny defenders access to capable open systems. Instead, openness must be paired with strong security safeguards, clear rules against misuse, rigorous evaluation and rapid remediation.^【10】