Welcome to fledd.io — a blog dedicated to the offensive security of Large Language Models.
This space explores the evolving attack surface of LLMs through empirical testing, adversarial prompt engineering, and systematic vulnerability analysis. From prompt injection and jailbreak persistence to latent behavior extraction and alignment circumvention, we document methods to probe, manipulate, and exploit LLM behaviors under various threat models.
Each post aims to contribute to a deeper technical understanding of how these models generalize, where their boundaries leak, and what that means for security-critical deployments in the wild.
Hands-on demonstrations and code samples are provided to illustrate attack vectors and surface misalignment phenomena — all intended for responsible research and defensive development.