Nvidia engineers detail Open Agent Safety Platform's three-layer approach to agent containment
LuizaJarovsky · x · 2026-11-01
AI governance researcher Luiza Jarovsky interviewed Ali Golshan, a senior engineer behind Nvidia's new Open Agent Safety Platform, covering how the application/runtime/infrastructure layers make policies verifiable and keep AI agents continuously contained and monitored. The discussion also explains why openness matters, with OpenShell released under Apache 2.0, framed against recent incidents like the OpenAI/Hugging Face event and AI-powered hacks.
Related event: Nvidia Unveils Open Agent Safety Platform for Verifiable AI Agents(2 posts)→
More from Models
- User's stale Chrome tab preserves ChatGPT's rolled-back model selector UI — Angaisb_ · 2026-11-01
- Dev: Opus 5.5 spoiled me, I hate talking to OpenAI models now — marclou · 2026-11-01
- Local STT shootout: Nemotron 3.5 Transcribe beats Fable on noisy audio — CryptoxPathy · 2026-11-01
- Mistral Large 4's Terminal-Bench 41% only holds inside Mistral's own agent — julsimon · 2026-11-01
- Mistral open-weights Large 4: ~1T-param MoE with 1M context, free until you host it — AIFlow_ML · 2026-11-01
- Anthropic to add invisible watermarks to Claude output worldwide starting Nov 3 — StewartalsopIII · 2026-11-01