NVIDIA unveils OpenShell, a hardware-backed safety platform with BlueField-4 as an external agent watchdog
mark_k · x · 2026-09-28
NVIDIA announced OpenShell, a hardware-backed safety platform for AI agents: each agent runs sandboxed with controlled access to files, networks, tools, and credentials.
On Vera Rubin systems, a BlueField-4 chip monitors agents from outside the host, sitting on the path to the model so it can enforce policy and interrupt an agent that goes off course — and the agent can't disable its own watchdog. The framing: containing capable agents is how we enable more autonomy.
More from Infra
- David Friedberg: Open-Source Models Make Data Centers Unnecessary for 90% of AI Use — MarvinTBaumann · 2026-09-28
- Old-School Infra Engineers on PB-Scale AI Logs: Hadoop Did This on a Tuesday — basedjensen · 2026-09-28
- Power sharing fits 192 GB300 GPUs in 140-GPU power budget, boosting tokens/s by 49% — julsimon · 2026-09-28
- Node.js async zstdCompress() was 23x slower due to missing pledged input size — AI Snake Oil · 2026-09-28
- Category theory meets deep learning: PyNCD diagrams derive hardware-aware FlashAttention — GioeleZardini · 2026-09-28
- NVIDIA adds $150B to buyback program, lifting total authorization to $235B — firstadopter · 2026-09-28