Building a Hackbot: How Ethiack's Hackian Finds Real Vulnerabilities
summary
Walks through an agentic pipeline that pairs LLMs with tooling (scrapers, FFUF-like fuzzing, CLI LLM connectors) and a four-part design: brain, structure, prompting, tools. Shows two autonomous findings: RCE via shell injection (breaking out of a grep context) and arbitrary file read via Clojure slurp/path input, both discovered in under four hours. Describes three-layer guardrails, per-target scope enforcement, and claims 99.5% precision, 200 vuln classes, 75% PortSwigger coverage.
AI-written summary · about
chapters (7)
- 0:00Introduction to hackbots
- 3:24The evolution of AI in security
- 7:20AI-powered hacking applications
- 11:18Designing a hackbot
- 17:56Implementing safety guardrails
- 21:20Show and tell: demonstration
- 26:32Human-AI collaboration
related talks
13mThe Automation of Exploits: Will Eatherton on AI-Driven Security and the End of Obscurity
38mHackers Gonna Hack: Identifying and Fixing Vulnerabilities Proactively
49mAutonomous Hacking Systems - Future Risk or FUD?
35mInside a hacker’s mind
31mA Dummy's Guide to Agentic Exploit Generation
39mNight of the living vulnerabilities
57mEnterprise Ethical Hacking Cloud Applications
41mLive Hacking: Breaking into Your Web App