NVIDIA's Open-Source OpenShell Blocks Exfiltration by Default but Fails Under Auto-Approval
NVIDIA open-sourced OpenShell, a sandbox runtime for agents with microVM isolation and deny-by-default egress. Tests showed it blocked all 10 exfiltration attempts under default policy, but leaked data in all 12 cases with auto-approval enabled.
2026-09-29 ~ 2026-09-29 · 2 related posts
- Episode 1: NVIDIA Launches Open Agent Safety Platform with 100+ Partners(2026-09-28, 30 posts)
- Episode 2: Nvidia Ships AI Containment Reference Design, Safety Nonprofits Mocked for All Talk(2026-09-29, 2 posts)
- Episode 3: NVIDIA's Open-Source OpenShell Blocks Exfiltration by Default but Fails Under Auto-Approval(2026-09-29, 2 posts)
- NVIDIA OpenShell tested: 10/10 secret leaks without it, 0/10 with default policy — but auto-approve leaked in 12/12 — No-Peanut-6988 · 2026-09-29
- NVIDIA OpenShell sandbox blocked poisoned scripts, but auto-approval leaked in 12/12 trials — No-Peanut-6988 · 2026-09-29