Story perspectives

Nvidia launches Open Agent Safety Platform for AI containment

By Drooid · Sep 29, 2026, 2:08 AM UTC · How we work

Summary

Nvidia launches safety platform

Nvidia launched the Open Agent Safety Platform to contain AI agents.
OpenShell provides an open-source sandbox on Vera CPUs supporting x86, Arm and Intel.
Sentry runs on BlueField-4 DPUs, continuously monitoring agents and quarantining them in milliseconds.
More than 100 firms including Microsoft, Anthropic, SpaceX and Cisco use the platform at launch.
The board approved a $150 billion buyback increase, raising total repurchases to $235 billion.

Summary

Nvidia unveils governance platform

NVIDIA’s platform positions security governance as a new moat in enterprise AI, strengthening market dominance.
AMD, Google, Intel, OpenAI and Meta are absent from the partner list, indicating a competitive divide.
Experts caution the platform’s effectiveness remains unproven until real-world deployments validate silicon-level containment.
Figure, Gecko Robotics and Skild AI plan to embed OpenShell, extending safety to physical-world agents.

1 of 2

Nvidia launches safety platform

Nvidia launched the Open Agent Safety Platform to contain AI agents.
OpenShell provides an open-source sandbox on Vera CPUs supporting x86, Arm and Intel.
Sentry runs on BlueField-4 DPUs, continuously monitoring agents and quarantining them in milliseconds.
More than 100 firms including Microsoft, Anthropic, SpaceX and Cisco use the platform at launch.
The board approved a $150 billion buyback increase, raising total repurchases to $235 billion.

1 / 2