Dev asks if their agentic upgrade safety plan covers the real attack surface

PitifulAppearance509 · reddit · 2026-11-22

A developer on Reddit is weighing whether to upgrade a simple RAG-based help assistant into a true agentic setup where the LLM queries the RAG, reads from the database, and potentially gains write access.

Worried about new attack vectors, they propose:

They ask the community whether this is on the right track or if key protections are missing.

Original post →

More from coding & agent

coding & agent channel →