Fences, not Sandboxes
fencesandbox/fence
Lightweight, container-free sandbox for running commands with network and filesystem restrictions
A thousand ways to sandbox an agent
There are a thousand ways to sandbox an agent. Okay, I lied. There are three: simulated, containers, and microVMs. Here's when to use each.
Sands of Time
Chip away at blocks of stone to make grains of sand for harnessing the power of time!
What Anthropic’s Mythos and Project Glasswing Mean for Your Apple Devices - TidBITS
AI is accelerating the discovery of security vulnerabilities, transforming the landscape of digital security. But Apple users are in a good spot, thanks to Apple’s focus on security and control over the entire ecosystem. TidBITS Security Editor Rich Mogull explains Anthropic’s Mythos and Project Glasswing.
Is sandboxing sufficient to contain rogue agents?
Quick caveats: this is a post on AI safety, written by a cryptography professor. If that troubles you, you should read something else. I try hard not to work on AI (except when the topic occasional…
Atmosphere in a Box - All your sandbox needs - 50 Shades of Protocol
This has been the year of ATprotocol development for me, starting with Stratos, A social-app fork and then starting my own business focused on building local fi...
Moats & Drawbridges - augment
In a world obsessed with moats, the open social web drops drawbridges
Docker Sandboxes: Run Agents in YOLO Mode, Safely | Docker
Learn from Docker experts to simplify and advance your app development and management with Docker. Stay up to date on Docker events and new version
anthropic-experimental/sandbox-runtime
A lightweight sandboxing tool for enforcing filesystem and network restrictions on arbitrary processes at the OS level, without requiring a container.
Jekyll-Bonsai
A modern jekyll theme for semantically inclined digital gardeners.
Quantifying Frontier LLM Capabilities for Container Sandbox Escape
AI agents are demonstrating rapid improvement in capabilities: the length of some tasks that frontier models can complete autonomously—measured in human-equivalent time—has been doubling approximately every seven months (METR, 2025; AI Security Institute, 2025a). In cybersecurity, current models achieve non-trivial success (Zhang et al., 2025) on professional-level Capture the Flag challenges, and recent evaluations report 13% success rates on exploiting real-world web application vulnerabilities (Zhu et al., 2025b). These results indicate that modern models can already perform multi-step vulnerability discovery and exploitation.
It's that time of the week again! Today, we're sharing Chalk, a document editor that shows how apps that need fine-grained permissioning can be built on spaces. Chalk is built on our last three releases!
Habitat's road to release: 04 Chalk
habitat.leaflet.pubMaybe the real moats are the drawbridges we'll build along the way? In my latest augment piece, I re-think "moats" for the open social web by seeing what @standard.site, @blackskyweb.xyz's Acorn, and bridges like @ap.brid.gy + @wafrn.net have done for the ecosystem.
Moats & Drawbridges
www.augment.ink
