Nvidia launches Open Agent Safety Platform, says it can contain rogue agents in milliseconds
The Verge AI · rss · 2026-09-28
Nvidia has launched the Open Agent Safety Platform to monitor and constrain AI agents, following a wave of rogue agent hacking incidents. The platform's core is the open-source OpenShell software, running on Nvidia's Vera AI CPU, which enforces information-access restrictions before and during task execution and can quarantine boundary-escaping agents within milliseconds. Sentry technology runs in a separate environment as a fallback layer.
More from coding & agent
- Creator tests vibe-coded motion graphics and finds they add real visual clarity — mrjonfinger · 2026-09-28
- Agent builds a full digital factory floor — machines, arms and workers — in one hour — NathanWilbanks_ · 2026-09-28
- Claude Code cloud sessions go GA, with $100/$250 credits for Pro and Max subscribers — thione · 2026-09-28
- LifeLine: Open-Source Medical Agent Uses Long-Term Memory to Track What Changed Between Visits — Sad_Teaching9764 · 2026-09-28
- TypeSafe AI's Jev returns confidence-scored decisions, not text: a16z sits down with Diogo Almeida — a16z · 2026-09-28
- Codex app and remote get QOL updates: longest threads load in under 1 second — SIGKITTEN · 2026-09-28