Comprehensive taxonomy of AI security vulnerabilities, LLM adversarial attacks, prompt injection techniques, and machine learning security research. Covers 71+ attack vectors including model poisoning, agentic AI exploits, and privacy breaches.

tracking AI system failures with solar weather

This repository documents a controlled research experiment that demonstrates how DNS lookups triggered by rendered content can be used to exfiltrate data. The technique leverages the browser's automatic DNS resolution behavior when rendering hostnames, without requiring HTTP requests or file upload