SUNGLASSES - AGENTIC AI SECURITY

SUNGLASSES is the open source input firewall for AI agents. It scans what your agent reads, like web pages, tool results, README files and MCP tool descriptions, and catches prompt injection, tool poisoning and agent supply chain attacks before the agent acts on them.

This channel shows those attacks in short episodes. Each one takes a real technique, shows how an AI agent gets tricked, and ends with the one habit that keeps it safe. Made for engineers and founders shipping AI agents, and for anyone curious how agent security really works.

What you get here
New shorts every day
Prompt injection explained with real examples
MCP tool poisoning, rug pulls and hidden instructions in files and web pages
Plain language, no hype

The scanner is free, local first and MIT licensed. 1583 patterns across 117 categories as of October 2026.

pip install sunglasses
sunglasses.dev
github.com/sunglasses-dev/sunglasses

Content reviewed by SUNGLASSES.dev.