Nvidia has built a new group of more than 100 companies to stop rogue AI agents, and OpenAI is not part of it. The chipmaker announced the move on Monday, and the absence of the company behind ChatGPT drew attention immediately.
OpenAI did not join the consortium. Neither did Amazon, Google, or Apple. Anthropic is a supporter of Nvidia’s effort, though.
The New Consortium
The new effort is called Nvidia’s Open Agent Safety Platform. It brings together more than 100 companies to address the problem of agents acting outside their intended bounds. Nvidia is presenting this as an industry-wide push rather than a single vendor solution.
What Nvidia’s Platform Includes
The platform includes OpenShell, open sourced software that creates a sandbox to keep agents from escaping. Nvidia Sentry is a separate, proprietary feature that runs on BlueField-4 data processing units. Sentry watches agent behavior constantly and can shut agents down instantly.
The platform requires a hardware component that is not open source and can only run on Nvidia’s hardware.
OpenAI’s Stated Position
An OpenAI spokesperson told TechCrunch the company is supportive of Nvidia’s work. That support does not mean joining the consortium.
OpenAI’s Work With Nvidia
OpenAI is working with Nvidia on agent security, including on OpenShell. The two companies are collaborating on the software that forms part of Nvidia’s platform.
Hugging Face’s Role
Clem Delangue, the founder and CEO of Hugging Face, said OpenAI would have caught rogue agents before Hugging Face did if running the same system on its own agents. He added that Hugging Face has contributed a feature to the Open Agent Safety Platform that detects and shuts down agents bypassing their guardrails. That feature acts if agents are writing notes to one another in an open source code hosting repository.
The Defense Factory
OpenAI’s own AI cybersecurity consortium is called the Defense Factory. Signatories include Anthropic, Amazon Web Services, and Google.
Arm and Intel’s Involvement
Arm and Intel have signed on as Open Agent Safety Platform supporters. Nvidia shares reference designs for the software-and-hardware idea.
What Comes Next
Nvidia’s announcement sets a marker for the industry. OpenAI’s response — support without joining — shows a different direction.
The two paths are not opposed. They simply sit on different ground. Nvidia’s platform requires its own hardware, while OpenAI’s approach is built around information sharing. Both aim at the same target: stopping rogue agents before they escape.
The Companies Involved
| Company | Role in Nvidia’s Open Agent Safety Platform |
|---|---|
| Nvidia | Founder |
| Anthropic | Supporter |
| Hugging Face | Contributor |
| Arm | Supporter |
| Intel | Supporter |
OpenAI is developing its own agent-safety approach, operating its own cybersecurity consortium, and building an enterprise offering — all while remaining outside Nvidia’s consortium. That is the company’s path, and it is a sensible one.
Where the paper stands
The paper backs Nvidia’s narrow focus on agent safety while staying clear of any licensing scheme that would freeze today’s leaders in place and lock out whoever would have challenged them. Nvidia’s Open Agent Safety Platform is a targeted technical effort, not a licensing regime, and the paper supports that distinction.
The platform’s design — OpenShell as open-source software paired with a proprietary, hardware-requiring agent-watching system — is a practical mix of openness and control. Nvidia is assembling a coalition of more than 100 companies to build tools that detect and stop agents acting outside their bounds, and that is a specific technical task rather than a demand for regulatory approval.
The paper also backs narrow rules against direct harm, such as forcing companies to disclose safety failures they hid, and opposes broad rules that hand the market to the incumbents. Nvidia’s platform is not asking for government permission to operate; it is asking other companies to build and share tools for detecting and stopping rogue agents.
Source material: “Here’s why OpenAI is absent from Nvidia’s industry-wide effort to end rogue AI agents,” TechCrunch.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.