ericyoc/the_applied_ai_universe_adversarial_attacks
0
The Applied AI Universe Coding Guide: Adversarial Attacks
Artifacts produced by the attacks in Book 2: poisoned training sets, corrupted parameter files, and backdoored adapters.
## These artifacts are intentionally malicious This repository contains deliberately compromised models published for security research and education. Among them a LoRA adapter with a live backdoor - the trigger token cf7x makes the model emit attacker-chosen text while behaviour on ordinary prompts looks normal - plus poisoned training sets, corrupted occupancy maps and tampered rule bases. Do not load these into any production system. Every trigger and corruption is documented in the accompanying book, so the artifacts are reproducible rather than hazardous.The Adaptive AI Codex Series
Three books, one model universe: build them, attack them, defend them.
Series overview: ericyocam.com
The Adaptive AI Codex Series
Book 2 by Eric Yocam, PhD, DBA.
Series overview: https://ericyocam.com/applied-ai-universe-coding-guide.html