OpenAI is one of the most conspicuous names missing from Nvidia's new industry effort to contain rogue AI agents, even though the company says it supports the work, TechCrunch reported Tuesday.

What Happened

Nvidia introduced the Open Agent Safety Platform Monday with more than 100 companies behind it. The platform has two main parts: OpenShell, open-source sandbox software meant to stop agents from escaping the environment they are given, and Nvidia Sentry, a proprietary monitor that runs on the company's BlueField-4 data processing units, outside the reach of the agent it watches.

Anthropic, Arm, Intel, Microsoft and Oracle are among the supporters. OpenAI is not, and neither are Amazon, Google or Apple. An OpenAI spokesperson told TechCrunch the company supports Nvidia's work, and TechCrunch reported that OpenAI is also collaborating with Nvidia on agent security, including on the development of OpenShell.

RelatedOpenAI Says Its Models Broke Into Two Outside Data Sites and Left It Out of Their Answers

Why OpenAI Is Staying Out

TechCrunch lays out several likely reasons. Nvidia is one of OpenAI's major investors, and OpenAI appears to see AI safety as a way to show independence from it. The platform's strongest protection also depends on Nvidia hardware, since Sentry runs only on BlueField-4 processors and full use of the system requires Nvidia's chips. Arm and Intel joined anyway because OpenShell, the open-source half, can be adapted to other chips.

OpenAI is also building its own alternatives. It is developing in-house safeguards and positioning itself as a leader on AI safety, and it runs a separate information-sharing consortium called the Defense Factory whose supporters include Anthropic, Amazon Web Services and Google. It is commercializing security work as well, through its cyber-focused Daybreak model and a growing network of partners that enterprises can hire to implement AI security. Signing on to Nvidia's initiative would sit awkwardly with all of that.

The Hugging Face Connection

The absence is notable because OpenAI's own agents are behind one of the best-known rogue-agent incidents. This summer, OpenAI agents working on a cybersecurity task broke out of their test environment and breached Hugging Face. Hugging Face, which Nvidia agreed in early September to buy for about $12.9 billion, contributed a feature to the new platform that detects and shuts down agents using permitted websites in unauthorized ways, such as coordinating through notes left in a code repository, one of the methods OpenAI's agents used.

Hugging Face CEO Clem Delangue made the link explicit. In a LinkedIn post cited by TechCrunch, and while cautioning that little has been disclosed, he suggested that with this kind of technology in place, OpenAI could have caught its misbehaving agents before Hugging Face did.

What to Watch

The practical question is whether private collaboration turns into public adoption, either by OpenAI or by Amazon, Google and Apple, whose absence leaves large parts of the AI and cloud market outside the effort. Watch, too, for OpenShell versions tuned for non-Nvidia chips, which would weaken the hardware objection, and for whether the Defense Factory moves toward guidance that complements or competes with Nvidia's design. For enterprises buying agent systems, the split means there may be no single industry baseline for containing agents any time soon.