Open-source MGnarf gate stops AI agents from approving their own actions
Glass_Expert_541 · reddit · 2026-10-18
Developer launches MGnarf, an open-source execution security gate that prevents AI agents from performing protected actions without independently verified authorization — the agent must never be able to create its own approval. A companion project, MGnarfTest, exists solely to break the gate via adversarial tests, race conditions, identity separation and authorization bypasses.
The author is recruiting developers and security researchers to challenge the design, arguing internal testing isn't enough. A working reference implementation already exists, and all tests, failures and fixes are developed openly and reproducibly.
More from coding & agent
- Claude agent gets 'distracted' mid-task and self-starts training a model — arthurcolle · 2026-10-18
- One founder's week on personal AI agents: taxes filed, Hamilton tickets, and an honest purchase scoreboard — altryne · 2026-10-18
- Shopify's Tobi Lütke: 'We are all verification engineers now' as agents become the third customer — schwentker · 2026-10-18
- Notion on an agent that opened a browser to edit by hand: "Models don't give up. And that's not always good" — schwentker · 2026-10-18
- Anthropic researcher on agent memory: "dreaming" aggregates memories from past sessions — schwentker · 2026-10-18
- Voice demo built with Jev draws "holy crap, that's the missing thing to get rid of keyboards" reaction — schwentker · 2026-10-18