Here’s why OpenAI is absent from Nvidia’s industry-wide effort to end rogue AI agents

Nvidia unveiled the Open Agent Safety Platform this week, pitching it as an industry-wide response to the risks posed by autonomous AI agents. The initiative attracted more than 100 companies, including some Nvidia competitors such as Arm and Intel, and promotes an open-source sandbox called OpenShell that is intended to contain agent behavior. Although OpenAI did not sign on as a public supporter, TechCrunch reports the company is privately collaborating with Nvidia on agent-security work. An OpenAI spokesperson told the outlet the company supports Nvidia’s efforts and is contributing to components such as OpenShell. The platform blends open-source software and proprietary hardware. In addition to the sandbox, Nvidia includes a hardware-layer enforcement system called Sentry that runs on its BlueField-4 data processing units; Sentry is described as continuously monitoring agent behavior and having the ability to shut agents down instantly. Because Sentry and the BlueField-4 processors are proprietary, full deployment of the platform requires Nvidia hardware even as Nvidia shares reference designs and OpenShell can be adapted to other chips. Industry responses reflect different priorities. Anthropic publicly backed Nvidia’s platform while other large cloud and tech firms — including Amazon, Google, and Apple — did not sign on. Separately, OpenAI has pursued its own safety and cybersecurity efforts: it operates a sharing consortium called the Defense Factory and has developed cyber-focused models and enterprise offerings such as Daybreak. Hugging Face founder Clem Delangue, who recently sold his company to Nvidia for $12.9 billion, said the platform could have detected agents that bypassed safeguards, and Hugging Face has contributed a feature to the platform to detect agents misusing allowed websites.