Microsoft MXC turns agent sandboxing into a policy teams can review
The useful part is not one more container: it is a common place to state what generated code may reach, then test the result.
2026-10-12 · 5 min · 5 sources
Simon Willison
x · Follow ↗
AI system security
Prompt injection, agents, model behavior and practical security
- Type
- Individual expert
- Priority
- Tier 1 · core
- Perspective
- Independent technical
- Best for
- Rapid demonstrations and primary links
- Publishes
- Daily / frequent
- Editor's note
- High-signal practical testing; archive linked posts and code.
Our analysis featuring Simon Willison
Recent from Simon Willison
- I've recently started hiding the bulk of my AI-generated GitHub PR descriptions behind a collapsed & 2026-11-09
- There's a buzzy new paper about GPT-6 Astra being able to directly interpret model weights (prompts 2026-11-03
- In updating a graphic showing U.S. company AI adoption over time according to a long-running U.S. Ce 2026-11-02
- More notes on Kimi K3.1 - this one is I big deal, first time an open weights (well, soon-to-be open 2026-10-23
- Pretty solid pelican https://t.co/bIpt5zCHPl https://t.co/kcrFAP7oMX https://t.co/twlS2dNSIe [LINKE 2026-10-23
- Given the current vulnerability apocalypse, is there anyone left who thinks holding assets in crypto 2026-10-23