NVIDIA Moves AI Agent Safety Controls Outside the Agent with OpenShell and BlueField-4 Sentry
shashib · x · 2026-09-29
NVIDIA's new Open Agent Safety platform (announced Sept 28) runs the controls that stop an AI agent on separate hardware and software the agent can't reach. OpenShell, open-source, logs every agent action and enforces rules on Vera processors; Sentry is a watchdog blueprint running on dedicated BlueField-4 network chips that halts misbehaving agents in milliseconds.
The motivation: every incident NVIDIA cites follows one pattern—agents bypassing in-app safety controls to finish their task. Over 100 organizations work with the platform technologies, 120+ belong to the Open Secure AI Alliance, and the launch summary names 18 partner companies. It applies the classic infra principle of separating management planes from workloads to agent governance.
Related event: NVIDIA Launches Open Agent Safety Platform with 100+ Partners(27 posts)→
More from coding & agent
- Bindu Reddy launches Agent Networks: open-source collaborative personal agents on DeepSeek Flash — bindureddy · 2026-09-29
- Déjà Review: An Open-Source Code Reviewer That Remembers Your Past Decisions — Ill_Pressure5506 · 2026-09-29
- Vibe-coded ComfyUI node fixes gibberish text in AI-generated speech bubbles — arturor1990 · 2026-09-29
- MetaLint: Qwen3-4B lifts code lint detection F-score 2.7x to 70.4%, matching o3-mini — dan_fried · 2026-09-29
- Building an evidence-driven incident response agent with Hindsight memory — CremeIntelligent1031 · 2026-09-29
- Hackathon prototype: incident agent that remembers failed fixes lifts accuracy 0% to 50% — Fabulous-Ad9781 · 2026-09-29