A field guide to sandboxes for AI

kept by eddie jan 07

Sandboxing tools for AI agents differ fundamentally by boundary (containers, gVisor, microVMs, Wasm), not just by policy, and picking the wrong one creates either leaks or excessive cost.

Two computers, one monitor, zero fiddling
A monitor with built-in KVM and DDC commands lets a user switch between a Mac laptop and Linux desktop with a single keyboard shortcut, eliminating cable fiddling.
also on KVM, #developer-tools
GitHub - kubernetes-sigs/agent-sandbox: agent-sandbox enables easy management of isolated, stateful, singleton workloads, ideal for use cases like AI agent runtimes.
Kubernetes SIG Apps introduces agent-sandbox, a CRD and controller for managing isolated, stateful, singleton workloads like AI agent runtimes.
also on gVisor, Kubernetes, #ai-agents, #developer-tools
How we contain Claude across products
Anthropic details how it contains AI agent blast radius across three products using sandboxes, VMs, and human-in-the-loop controls, while documenting missed risks like pre-trust code execution and user-mediated prompt injection.
also on gVisor, #security, #ai-agents, #developer-tools
$1 million hacker challenge for Vercel Sandbox - Vercel
Vercel launches a two‑week public HackerOne bounty offering up to $1 million for researchers who can escape its Sandbox microVM isolation.
also on Firecracker, #security, #developer-tools
Deno Sandbox | Hacker News
Deno Sandbox launches microVMs that control network egress and protect API keys from exfiltration in LLM-generated code, then deploy directly to Deno Deploy.
also on Firecracker, #security, #developer-tools