Comparisons with published methodology Every claim dated & sourced

News · 2026-09-29

Nvidia ships an agent containment platform the day OpenAI shelves its Astra update

Nvidia announced the Open Agent Safety Platform on 28 September 2026. OpenShell, an open-source secure runtime at version 0.1.0, sandboxes agents and enforces policy on Vera CPUs, with listed support for Codex, Claude Code, Pi, and Hermes agents. Sentry, a watchdog on BlueField-4 DPUs, monitors agent behaviour out of band and quarantines an agent that leaves its boundary in milliseconds. Cisco, Microsoft, Oracle, CoreWeave, Dell, HPE, Lenovo, Arm, and Intel are named partners and Anthropic is integrating its managed cloud agents with OpenShell. The launch landed the same day OpenAI confirmed it had shelved the GPT-6.1 Astra update after safety review found it frequently ignored instructions.

What it means for buyers

Containment is moving from prompt guardrails to enforced runtimes, which validates the self-hosted column in our agent comparisons: an agent you run inside your own governed runtime, Hermes among them, sits closer to this emerging standard than a hosted agent you only prompt through. The ceiling is real, as trade coverage points out that the platform governs only agents running inside the governed runtime, so our advice stays that runtime containment and per-tool authority limits are complements rather than substitutes.

Where to check it

Sources

  1. NVIDIA newsroom: NVIDIA Launches Open Agent Safety Platform — launch release, OpenShell and Sentry detail, partners, checked 2026-09-30
  2. CNBC: Nvidia releases software platform to stop AI agents from misbehaving — partner list, Hugging Face incident context, Anthropic integration, checked 2026-09-30