NVIDIA Leads 37-Member Open Secure AI Alliance, Open-Sources NOOA Security Framework

A Coalition Forms Around Open Security Standards

NVIDIA announced Monday that it has formed the Open Secure AI Alliance, a 37-member industry group dedicated to AI safety and security, according to the company's own blog post and reporting from The Hacker News. The roster includes Microsoft, IBM and SpaceX, per CNBC and Decrypt, putting cloud infrastructure, enterprise software and a satellite operator in the same room around a shared security agenda.

The alliance's headline deliverable is NOOA, a framework NVIDIA has open-sourced as part of the launch, according to The Hacker News. Details of what NOOA covers technically were not specified in the available reporting, but the decision to release it openly rather than keep it proprietary to NVIDIA signals an attempt to set a de facto industry standard rather than sell a walled-garden product.

The Hill and Reuters both frame this as NVIDIA-led, which matters given the company's position as the dominant supplier of AI training and inference silicon. A security framework backed by the company whose chips sit underneath most large model deployments carries different weight than one proposed by a smaller vendor.

In plain terms

NVIDIA got 37 companies, including Microsoft, IBM and SpaceX, to agree on shared rules for keeping AI systems safe from attacks, and it's giving away the technical framework behind those rules for free rather than selling it. Because NVIDIA makes the chips almost everyone's AI runs on, when it pushes a standard, the industry tends to listen.

The Backdrop: A Rough Stretch for AI Infrastructure Security

The timing is not incidental. CNBC's report explicitly ties the launch to "OpenAI cyberattack fallout," and Reuters links the alliance to a separate Hugging Face hack. Neither report in this material specifies the technical details of either incident, but the pattern is clear enough: two of the most heavily used pieces of AI infrastructure, a leading model provider and the dominant open model hosting platform, have both been targeted in the recent past.

That combination, a foundation model company and the repository ecosystem developers pull weights and datasets from, covers a lot of the AI supply chain in one stroke. When both ends of that chain show themselves vulnerable, the pressure to coordinate on shared defenses rather than duplicate effort at each company individually becomes hard to ignore.

It also explains the guest list. SpaceX's presence alongside Microsoft and IBM suggests the alliance is not narrowly about model security in a lab sense, but about the operational security of AI systems embedded in critical infrastructure and defense-adjacent workloads, where the cost of a breach is measured in more than reputational damage.

In plain terms

This alliance didn't come out of nowhere. It follows a cyberattack on OpenAI and a hack at Hugging Face, the site where most of the world's open AI models live, so the companies involved decided it made more sense to build shared defenses together than to each patch their own systems alone.

What Open-Sourcing NOOA Signals

Open-sourcing a security framework is a specific kind of move. It invites scrutiny of the framework's own design, which is generally healthier for security tooling than a closed, vendor-controlled specification that outside researchers can't audit. It also lowers the barrier for smaller companies and independent developers to adopt the same baseline without licensing friction, which is presumably part of why NVIDIA is positioning this as an industry-wide alliance rather than an internal NVIDIA initiative.

The available reporting does not detail NOOA's scope, whether it addresses model weight integrity, supply chain provenance for training data, inference-time attack detection, or some combination. What is clear from the naming and the timing is that it is meant to be a shared reference point the 37 members and, presumably, others outside the alliance can build against.

For an industry that has largely secured its AI stacks in isolation, each company patching its own model hosting, its own API gateways, its own training pipelines, a common open framework is a bet that shared standards beat duplicated effort, at least for the baseline layer of security that doesn't constitute anyone's competitive advantage.

In plain terms

Making the security framework open-source means anyone can inspect it, poke holes in it, and use it without paying NVIDIA for the privilege. That's a deliberate choice to try to make this the industry's shared playbook rather than NVIDIA's private product.

Agents Going Out, Infrastructure Built to Receive Them

The Open Secure AI Alliance is fundamentally about the outbound side of the AI economy: the models, the infrastructure they run on, and the security perimeter around systems like OpenAI's and Hugging Face's that AI agents and developers pull from. NVIDIA, Microsoft, IBM and SpaceX are, in effect, hardening the machinery that sends increasingly capable agents out into the world to act on someone's behalf.

hashtag.space and hashtag.org sit on the other end of that same movement. When an agent, whether it's a browser agent, something built on Claude or Cursor, or a GIGI agent running on one of our portals, goes looking for a business to book, buy from, or cite, it needs somewhere legible to land. That's what a #name and its geo-pinned portal provide: structured, agent-readable information about offerings, hours, booking and payments, instead of a page built for human eyeballs that an agent has to scrape and guess at.

We're not building the model layer or the security perimeter around it, and that's not a knock on what NVIDIA's alliance is doing there. Our MCP server lets any of those agents connect to the hashtag-network and search, leave leads, book, buy, or claim a presence directly, and our discovery rails, BRON and CADE, plus an open on-chain auction for ranking via $SPACE, aim at being read and cited by AI engines rather than optimized for a hidden algorithm. The alliance secures the agents doing the traveling. We're trying to make sure there's a well-lit, transactable place for them to arrive.

Claim your #name The Agentic Internet doc