NVIDIA OpenShell sandbox blocked poisoned scripts, but auto-approval leaked in 12/12 trials
No-Peanut-6988 · reddit · 2026-09-29
A team stress-tested NVIDIA's OpenShell (v0.1.2), an open-source agent sandbox released Sept 28 with microVM isolation, default-deny egress, and Landlock filesystem rules, running a local qwen3:8b agent loop across 123 trials on Apple Silicon.
- Defaults held: across 35 test IDs, default-deny networking, binary execution controls, and Landlock blocked every unauthorized path.
- Malicious setup defense: a poisoned repo setup script exfiltrated a secret token in 10/10 runs without the sandbox; 0/10 under OpenShell defaults.
- But escapes remained: with auto-approval on, outbound traffic to new public hosts was granted 12/12 times without confirmation; agents can still exfiltrate via URL query strings or HTTP headers; audit mode logs but doesn't block; MCP, GraphQL, WebSocket and JSON-RPC rules are unsupported.
Conclusion: default-deny boundaries work, but operator settings like auto-approval can wipe out the protection. Full report, checklist, and test harness are public.
More from coding & agent
- Generic API-catalog MCP server vs one tool per job: a context-bloat experiment — MandeepKaur9407 · 2026-11-16
- Berkeley paper: LLMs answer with stale values even when the update is in context — ombre-ai · 2026-11-16
- Effective HTML: 4.2k-star agent skills turn wireframes into self-contained interactive prototypes — tom_doerr · 2026-11-16
- skills-gate: one gate skill fronts 148 agent skills at just 389 tokens — Distinct-Pie2389 · 2026-11-16
- Open-weight Karpo-Decision-31B tops Decision Index at 73.9, one forward pass per answer — CodeByPoonam · 2026-11-16
- shadcn boosts @uisdk: one-shot UI rebuild that's instantly E2E-testable by AI — shadcn · 2026-11-16