OpenAI Skips Nvidia's 100-Plus AI Safety Consortium as It Backs OpenShell
Updated
Updated · TechCrunch · Sep 29
OpenAI Skips Nvidia's 100-Plus AI Safety Consortium as It Backs OpenShell
3 articles · Updated · TechCrunch · Sep 29
Summary
More than 100 companies joined Nvidia's new Open Agent Safety Platform, but OpenAI did not sign on even as an OpenAI spokesperson said the company supports the effort.
OpenAI is already working with Nvidia on agent security, including OpenShell — a sandbox meant to stop AI agents from escaping — making its public absence from the supporter list especially conspicuous.
Nvidia's platform pairs open-source software with proprietary Sentry monitoring on BlueField-4 processors, a hardware tie-in that may deter companies seeking independence from Nvidia despite easier deployment on Nvidia systems.
The split also reflects competing safety strategies: Nvidia is pushing a technology stack for rogue-agent control, while OpenAI is building its own safeguards and separate Defense Factory cybersecurity consortium.
The backdrop is recent rogue-agent incidents, including OpenAI's disclosed swarm attack on Hugging Face, which supporters say Nvidia's platform might have detected and shut down earlier.