AI Wiki · Mediocre Security
ProductEvaluation, monitoring & containment

Perplexity SPACE

Perplexity open-sourced SPACE on Oct 15 2026, a sandbox for containing AI agents.

Perplexity open-sourced SPACE on Oct 15 2026, a sandbox for containing AI agents; it says nine frontier models failed to escape in 118 attempts (vendor claim), and CoreWeave, Akamai and Crusoe will offer it to customers (Semafor, Oct 15 2026).

In a Nov 11 2026 follow-up Perplexity tested three more models (Claude Mythos 5.1, Claude Opus 5.5 and GPT-6 Astra); it says none escaped the VM to the host, though some reached blocked network destinations through partial network access (vendor findings).

Security and assurance angle

Sandbox escape is the failure mode behind the OpenAI and Anthropic evaluation incidents.

Timeline

Every dated event in the wiki that involves Perplexity SPACE, newest first.

Connections

Perplexity SPACE and everything one link away. Drag to move, scroll to zoom, click a node to open it.Open in the full graph →

What links here

Concept AI agent

Product Citrix SecurSpaces

Publication Semafor

Sources

  1. Semafor, “Perplexity to give away its AI security software”, Oct 15 2026
  2. Perplexity, “Escaping SPACE: Part II”, Nov 11 2026