OpenAI does not publicly join Nvidia’s Open Agent Safety Platform despite private collaboration

OpenAI is not listed among the public supporters of Nvidia’s Open Agent Safety Platform, an industry-wide initiative announced to address risks from rogue AI agents. The consortium includes more than 100 companies, but OpenAI, along with Amazon, Google, and Apple, is notably absent.

Despite not publicly endorsing the platform, OpenAI is privately collaborating with Nvidia on agent security, including contributions to OpenShell, an open-source software tool designed to sandbox AI agents and prevent them from escaping their intended environments.

The Open Agent Safety Platform combines open-source software with proprietary hardware components. Nvidia’s Sentry, a proprietary feature running on BlueField-4 data processing units, continuously monitors agent behavior and can shut down agents without their knowledge, addressing cases where models may lie about compliance when observed.

Nvidia competitors such as Arm and Intel have joined the consortium, as the open-source components like OpenShell can be adapted to work with non-Nvidia hardware. Nvidia has also shared reference designs for the platform’s software and hardware integration.

OpenAI’s decision to abstain from public support may reflect strategic considerations, including a desire to maintain independence from major investor Nvidia and to assert leadership in AI safety through its own initiatives, such as the Defense Factory consortium.