OpenAI did not become a public participant in Nvidia’s agent safety platform — TechCrunch
OpenAI did not become a public participant in the Open Agent Safety Platform, an Nvidia consortium joined by more than 100 companies to develop tools to counter uncontrolled behavior by AI agents. At the same time, an OpenAI spokesperson told TechCrunch that the company supports Nvidia’s work in this area and is collaborating with it on agent safety.
Companies that also have not publicly joined the initiative include Amazon, Google, and Apple. Anthropic, however, has supported it. Nvidia presents the platform as a way to spread its own, mostly open tools for protecting AI agents across the broader technology industry.
Collaboration around OpenShell
OpenAI is working with Nvidia, including on OpenShell, open-source software for creating an isolated environment intended to prevent AI agents from going beyond established boundaries.
More current news is available on the UA.News Telegram channel Telegram.
Hugging Face founder and CEO Clem Delangue said that using such a tool could have helped identify earlier an incident involving OpenAI agents that the company had reported previously. According to him, Hugging Face has already added a feature to the platform that detects and stops agents using permitted websites in an unauthorized manner. In particular, it is designed to respond to agents coordinating their actions through notes in open-source repositories.
Hardware-level protection
The full Open Agent Safety Platform includes not only OpenShell but also Nvidia’s proprietary Sentry feature for hardware-level monitoring. It runs on specialized BlueField-4 data processing units and, according to Nvidia, continuously tracks agent behavior and can immediately halt their operation.
This part of the system is not open-source software and is available only on Nvidia hardware. At the same time, Arm and Intel supported the platform because OpenShell can be adapted for other chips, while Nvidia distributes reference designs for the hardware-software architecture. OpenAI is simultaneously developing its own protective tools, as well as the Defense Factory consortium for sharing cybersecurity information, which has been supported by Anthropic, Amazon Web Services, and Google.
