Perplexity red-teams SPACE with frontier models: 108 escape attempts, zero host escapes
perplexity_ai · x · 2026-10-16
Perplexity's Escaling SPACE: Part I research threw nine frontier-model configurations, each with root inside a SPACE microVM, at the sandbox platform: 108 escape attempts, none reached the host.
The defense model: agents keep full root freedom inside the sandbox while the host stays out of guest control, with a per-sandbox gateway enforcing egress policy on all outbound traffic and credentials managed outside the VM.
Related event: Perplexity Open-Sources SPACE, Its Agent Sandbox Platform(7 posts)→
More from coding & agent
- Vercel Goes Full Backend: Python GA, Containers, and Daemons — cramforce · 2026-10-16
- WorldMemArena Accepted at NeurIPS: Memory Storage ≠ Memory Use — xwang_lk · 2026-10-16
- Vercel Ship Demos Container Deploys for Go, Ruby, Java and More — edwin · 2026-10-16
- TRL adds day-0 support for fine-tuning Muse Spark Open's 816B MoE with LoRA on 64 H100s — QGallouedec · 2026-10-16
- Databricks engineers land impactful PRs on day 2 with AI dev tools, vs week 3 before — Yuchenj_UW · 2026-10-16
- Perplexity open-sources SPACE, the sandbox platform powering its Computer agents — AravSrinivas · 2026-10-16