37 companies have lined up behind Nvidia in a new Nvidia AI security alliance, but OpenAI, Anthropic and Google are missing from the inaugural roster.
The group, called the Open Secure AI Alliance, launched Monday to build open-source security tools for AI systems, according to CoinDesk. Its core claim is blunt: cyber defenders need AI tools they can inspect, adapt and run on their own infrastructure when incidents are live.
Nvidia’s 37-member AI security alliance starts with one missing trio
The Open Secure AI Alliance includes Microsoft, IBM, Red Hat, Cloudflare, CrowdStrike, Palantir, Databricks, Hugging Face, SpaceXAI and the Linux Foundation, among others. Nvidia said the effort builds on the Linux Foundation’s Akrites initiative and OpenSSF work.
The headline absence is just as important as the membership list. OpenAI, Anthropic and Google, three of the most visible closed-model AI developers, are not listed as inaugural partners.
That does not prove opposition. The announcement does not say why they are absent. But it does put a hard edge on an already visible split: some vendors want security teams to rely more heavily on open systems they can control directly, while the largest model labs have centered much of their commercial AI around proprietary systems and hosted access.
Nvidia’s stated argument comes from a recent Hugging Face breach. In Nvidia’s account, the incident showed how difficult live response can become when defenders cannot inspect or adapt the AI systems they are using.
According to Nvidia, closed AI tools, “unable to distinguish attackers from defenders,” blocked forensic analysis. Hugging Face then ran GLM 5.2, an open-weight model from Chinese developer Z.ai, on its own infrastructure to analyze more than 17,000 actions and contain the intrusion.
“When defenders cannot inspect, adapt and run advanced AI on their own infrastructure, their ability to respond is constrained at exactly the moment speed matters most,” Nvidia said.
The self-run AI pitch is about incident response speed, not ideology alone
The alliance is selling a practical security workflow: when a breach is underway, defenders need AI that can review logs, trace agent behavior, inspect actions and work inside controlled environments without being blocked by outside systems.
Alliance materials describe the work as a shared push to develop open security components rather than a single hosted product. Based on the supplied source material, the clearest supported details are the roster, the open-source security focus and the emphasis on tools defenders can inspect, adapt and run themselves.
The announcement points broadly to work around agent behavior, auditability, model handling, supply-chain security and incident response. However, the precise status and purpose of several named member tools are not detailed in the supplied material, so the stronger claim is about the alliance’s intended stack rather than any one contribution.
That makes the Nvidia AI security alliance more than a branding exercise, at least on paper. It is trying to define a stack for AI defense: models, harnesses, identity, isolation, logs, evaluation and secure coding workflows.
Nvidia’s blog frames the issue as one of control. The company says open tools can help defenders protect data, customize controls and avoid single points of failure. It also acknowledges the risk that open models can be misused, including by attempts to weaken safeguards or repurpose capabilities for cyberattacks.
The balance matters. Open models give defenders more room to inspect and adapt systems. They also give attackers more material to study. Nvidia’s answer is not to deny that risk, but to argue that closed weights do not stop determined attackers either.
OpenAI, Anthropic and Google absence turns this into an enterprise AI control fight
The missing names sharpen the market signal. OpenAI, Anthropic and Google are not minor omissions. They are central to the closed-model side of the AI market, and their absence leaves Nvidia and its partners defining the alliance around a different operating model.
The Open Secure AI Alliance is not saying closed models have no role. Nvidia’s blog says the world needs both closed and open models. Its sharper claim is that cybersecurity requires open models and open harnesses because defenders need transparency, local control and the ability to run tools where sensitive work happens.
That framing also lands in a broader fight over AI control points. XOOMAR has tracked related pressure around platform power in Google Search ranking scrutiny and AI distribution in Stripe and OpenRouter talks. Those are separate stories, but they point to the same strategic question: who sits between users, models and critical workflows?
Here, the workflow is security. That makes the stakes less abstract. If a defensive AI system blocks analysis during an active incident because it cannot tell the responder from the attacker, the model’s safety layer becomes part of the operational problem.
For Nvidia, the alliance also fits its strongest position in AI: the infrastructure layer. The source material supports a narrower reading than “Nvidia versus the model labs,” but the business logic is clear enough as analysis. If more enterprises decide AI security tools must run under their own control, demand shifts toward systems that can be deployed, tested and governed inside private environments.
Crypto attacks give the alliance a high-pressure test case
The crypto angle is not incidental. CoinDesk notes that four protocols were drained of more than $35 million in a single stretch last week, including AFX, Verus and Bitcoin scaling network B².
None of those attacks broke cryptography, according to the source material. Each abused a trusted control. That is exactly the kind of long-horizon, multi-step work where AI systems are getting better, and where defenders may want fast analysis without handing sensitive details to an outside service.
A drained contract also differs from a corporate breach in one brutal way: it cannot simply be reversed. That raises the pressure on detection, response and pre-deployment review.
For the Nvidia AI security alliance, crypto offers a proving ground with visible consequences. If open agent tools can find exploitable bugs, audit agent behavior or speed forensic analysis before funds move, the value case becomes easier to understand.
The next test is deployment, not the launch list
The alliance now has names, tools and a clear narrative. What it does not yet have, based on the supplied material, is public evidence of broad customer deployment or measurable security outcomes from the new group.
The next proof points are concrete:
- Reference architectures: Security teams will need repeatable ways to run these tools safely.
- Evaluation standards: Buyers will want tests for model behavior, agent permissions and failure modes.
- Incident workflows: The alliance needs to show how open AI fits into live response without creating new risks.
- More members: If major model providers or additional security vendors join later, the group becomes harder to dismiss.
The practical watch item is whether the Open Secure AI Alliance turns its early open-security work into a usable deployment path for real security teams. If it does, Nvidia will have helped move AI security from hosted-model access toward controlled, inspectable systems that defenders can run when time is short.
Impact Analysis
- The alliance signals growing demand for AI security tools that organizations can control during live cyber incidents.
- The absence of OpenAI, Anthropic and Google highlights a widening divide between open and proprietary AI security approaches.
- Nvidia is positioning itself at the center of AI infrastructure, security tooling and open-source ecosystem development.
Originally published on XOOMAR. For more news and analysis, visit XOOMAR.
Top comments (0)