Prompt Engineering for Cyber Security
Workshop materials teaching security practitioners how to write, refine and verify prompts for real security tasks, including a hands-on prompt-injection challenge and packaging prompts into reusable Modelfiles.
This is a four-hour, hands-on workshop on getting professional-grade output from AI on real security tasks, teaching two layers of prompting most people blur together: building a strong prompt before the first draft, then running a refine-and-verify loop afterward. It is the second in a two-part series that follows a workshop on setting up local LLMs with Ollama, mesh networking and authentication, though a ready-to-go cloud VM is provided for those who skip the first session.
It covers building prompts with the right role, context, constraints and output shape; running a disciplined refine-and-verify loop on findings, configuration analysis, detections and proof-of-concept tooling; detecting and removing AI writing tells from client-facing text; packaging repeatable workflows into Skills or Modelfiles; building lightweight evaluation sets to compare prompt revisions; and crafting and recognizing prompt injection, jailbreaking and data-extraction attacks, including a hands-on injection challenge showing why prompt-based guardrails are not a security boundary.
Aimed at red teamers, penetration testers and other security practitioners; presented at the Antisyphon AI Summit in August 2026.