Nvidia backs open AI security alliance with tech giants

Nvidia and a group of technology companies have launched the Open Secure AI Alliance, which brings together companies and open source groups to develop shared AI security tools.

The group includes Nvidia, Adobe, Cisco, Cloudflare, CrowdStrike, Hugging Face, IBM, Microsoft, Red Hat, Salesforce, SAP, ServiceNow and Siemens, alongside the Linux Foundation and other partners from cloud computing, enterprise software and AI research.

It will focus on open technologies for identifying, fixing and disclosing vulnerabilities in software and AI agents. Its formation reflects a wider debate in the technology industry over whether cyber defence tools should rely mainly on closed systems from a small number of providers or on open models and software that users can inspect and adapt.

Nvidia said cybersecurity has become one of the main areas shaped by open source software, which already underpins sectors including cloud computing, finance, manufacturing, telecommunications and government systems. It argued that open approaches can help defenders avoid relying on a single supplier and give them more control over how tools are used on their own infrastructure.

Open stack

The alliance will build on existing work from the Linux Foundation's Akrites initiative and the Open Source Security Foundation community. Members want to create what they describe as an open defence stack for AI agents, covering identity systems, isolation, safe model formats, scanning tools and secure coding workflows.

Nvidia is contributing open models, model weights, data and research on agent harnesses. It has also released an open source project called Nvidia Labs Object-Oriented Agent, or NOOA, designed to make AI agent behaviour easier to test, trace, audit and govern.

Nvidia said AI safety depends not just on the model itself but on the broader software stack around it, including permissions, harnesses, guardrails, logs and evaluation systems. That framing matters because much of the public policy debate around AI safety has centred on whether model weights should remain closed.

Several partners highlighted existing projects that will feed into the alliance. HPE contributes to SPIFFE and SPIRE, standards used to verify the identity of workloads and services. Hugging Face has offered Safetensors, a format for storing model weights intended to prevent remote code execution, to the PyTorch Foundation.

IBM and Red Hat are bringing work from Lightwell, which uses digitally signed patches to strengthen software supply chain security. Microsoft's MDASH system uses multiple AI agents to scan for exploitable software bugs, while SpaceXAI has open-sourced a terminal-based AI coding agent known as Grok Build.

Policy debate

The launch comes as governments and regulators weigh how to handle increasingly capable AI systems. Supporters of open models argue that restrictions could leave defenders dependent on a handful of closed providers and limit their ability to inspect or run tools in sensitive environments.

Critics argue that open models can be modified or misused for cyber attacks. The alliance said those risks are real but not unique to open systems, and that they should be addressed through safeguards, evaluation and rapid remediation rather than blanket limits on access.

Nvidia pointed to a recent security incident at Hugging Face as an example. During the incident, the company used an open-weight model running on its own infrastructure to analyse more than 17,000 actions and help contain the intrusion after closed AI tools blocked parts of the forensic analysis.

Advocates of open AI security tools have cited the episode as a practical example of the challenge facing cyber response teams. If defenders cannot inspect, adapt or deploy advanced AI systems on their own infrastructure, they may lose critical time during an incident.

Broad backing

The list of inaugural partners spans a wide range of companies, including Capital One, Cloudera, Databricks, Dell Technologies, DoorDash, Elastic, LangChain, NAVER, NetApp, Nous Research, OpenClaw, Palantir, Palo Alto Networks, Reflection AI, SK Telecom, Snowflake, Synopsys, Thinking Machines Lab and TrendAI. The breadth of the group suggests the alliance is trying to reach beyond chipmakers and model developers into the wider software and cyber ecosystem.

The initiative also reflects a growing push by large technology companies to shape AI governance through open source structures rather than relying solely on proprietary products or regulation. By tying AI security to established open source institutions, the alliance is seeking to position openness as part of cyber resilience rather than a risk in itself.

Nvidia said companies and governments should invest in shared infrastructure for AI defence, including datasets, evaluation frameworks, attack simulators and red-teaming tools. It argued that "The right response is not to deny defenders access to capable open systems. It is to pair openness with strong safeguards, clear rules against malicious misuse, rigorous evaluation and rapid remediation."