become an AI HACKER (it's easier than you think)

become an AI HACKER (it's easier than you think)

by NetworkChuck

Applied Sciences & Engineering Computing & Cybersecurity URComputer securityURHComputer fraud and hacking
🎙 NetworkChuck 👥 5.4M 📅 February 20, 2026 ⏱ 16 min 👁 395K 📄 tutorial 🧭 2026-09-09
Available in: English (current) Français

Keywords

AI hackingAgent BreakerAuto Parts CTFprompt injectionLLM security

Summary

This video, hosted by NetworkChuck with guest expert Jason Haddix, continues a series on AI hacking. It bridges the gap between simple CTF challenges like Baby Gandalf and real-world AI penetration testing. The host first introduces Agent Breaker, a new Lakera lab that simulates hacking AI-powered applications with objectives such as manipulating risk ratings. He emphasizes the non-deterministic nature of LLMs, explaining why repeated attempts are necessary. The video then presents the Auto Parts CTF, a self-hostable Docker-based lab that mimics a real client engagement. Jason demonstrates how to leak system prompts, extract API keys, and expose sensitive patent data from a retrieval-augmented generation (RAG) database. The content includes a roadmap for beginners: start with Gandalf, move to Agent Breaker, then tackle the Auto Parts CTF, and finally explore additional labs and bug bounties. The host shares that a 12-year-old solved the CTF in 35 minutes, underscoring accessibility. The video also includes a sponsorship segment for Bitdefender and ends with a prayer, which has become a signature element. Overall, it provides practical resources and a clear learning path for aspiring AI security professionals.

183 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video offers high practical value by providing free, open-source labs and a self-hostable CTF, enabling hands-on learning. Argumentation is solid: the host demonstrates real attempts, including 239 trials on one level, illustrating the persistence needed. Jason Haddix’s expertise lends credibility, and the demonstration of leaking secrets from a realistic application makes the threat concrete. The step-by-step instruction is clear, though some segments feel repetitive. The use of a 12-year-old as an example effectively argues that the barrier to entry is low. The argument that AI hacking is a growing field with bug bounties and job opportunities is compelling, though it relies on anecdotal evidence rather than statistics.

Scientific Rigor, Source Quality, Title Accuracy

The video cites specific, verifiable sources: Lakera’s Gandalf and Agent Breaker, the Auto Parts CTF GitHub repository, an AI security resource hub, and course links. These are directly relevant and credible. The title is accurate and not sensationalized; it promises a learning path and delivers. Scientific rigor is acceptable for a tutorial, though some claims (e.g., ‘a 12-year-old solved it’) are anecdotal. The mention of bug bounties is accurate but not detailed. No sources are fabricated. Comments indicate a generally positive reception, with many viewers expressing appreciation for the prayer and the practical approach, though a few technical criticisms were noted.

224 words

Title / Content Match

The title accurately reflects the video's aim to teach AI hacking in an accessible way, emphasizing that 'it's easier than you think' matches the encouraging and progressive approach demonstrated. Content aligns well with the promise of practical, actionable steps.

Quality & Reliability

8/10

The video features Jason Haddix, a recognized AI pentesting methodology author, and uses verifiable, open-source resources (GitHub, Lakera). Technical claims are grounded in practical demonstrations and references to real-world engagements. The sponsorship segment is clearly separated and does not undermine the core educational content.

Chapters

Cited Sources

  • Baby Gandalf — Introductory AI hacking CTF challenge from Lakera, used as a starting point.
  • Agent Breaker — Advanced CTF simulating real AI application attacks, featured in the video.
  • Auto Parts CTF — Self-hostable CTF based on a real client engagement, used to teach real-world AI pentesting.
  • AI Security Resource Hub — Curated list of 23 active labs and resources for AI hacking practice.
  • Arcanum Security Courses — Official courses by Jason Haddix for advanced AI security training.
  • Part 1 — Hacking AI is TOO EASY — Predecessor video introducing Baby Gandalf and basic AI hacking concepts.

External References

Contribution & Novelties

The video’s original contribution lies in making real-world AI pentesting accessible through free, open-source resources and a self-hostable CTF that mirrors actual client engagements. It bridges the gap between toy challenges and professional practice, offering a clear roadmap for beginners. The inclusion of Jason Haddix provides authoritative insight into methodology that is not commonly available. The emphasis on persistence and the non-deterministic nature of LLMs is a valuable practical lesson.

Pour aller plus loin :

112 words

Radar Profile

The radar profile shows balanced scores across information quantity, quality, technical level, and reliability, with reliability and quantity slightly higher. This indicates a well-researched, resource-rich tutorial that is both practical and trustworthy, though technical depth is moderate.

Reliability 8/10

💬 Très positif — Sur les 30 commentaires analysés, la grande majorité exprime une gratitude sincère pour la prière finale, et plusieurs louent la clarté pédagogique et l'accessibilité du contenu technique. Quelques commentaires signalent un lien GitHub cassé, mais dans l'ensemble le climat est extrêmement favorable et empreint d'émotion.