Skip to content
skifo/
FR

AI Security — Attacks and Defenses

Two intensive days to understand, reproduce and counter attacks on AI models — both CNNs and LLMs — in industrial and adversarial settings.

2 days · 16×45min · isolated labs

All training Advanced course
Duration
2 days (16×45min)
Audience
ML / security engineers
Price
CHF 3,800/person or 24,000 team
Format
Lectures + flipped + labs

Who it's for

  • ML / MLOps engineers shipping models to production
  • Data scientists facing adversarial environments
  • Technical security teams auditing AI systems
  • Architects integrating AI into sensitive products

Learning objectives

  1. Identify and assess AI risks on CNNs and LLMs in industrial and adversarial settings
  2. Understand and reproduce adversarial examples, data poisoning, backdoors, LLM jailbreaks (white-box and black-box)
  3. Apply explainability techniques to audit and challenge AI decisions

Prerequisites

Fluent Python, machine learning fundamentals, understanding of neural networks (CNN, transformer). Labs assume the ability to read and modify code.

Teaching format

A combination of technical lectures, flipped classroom (prepared upstream then presented in groups) and hands-on labs in a provided isolated environment.

Workload
Format Duration
Lectures 8×45 min
Flipped classroom 4×45 min
Hands-on labs 4×45 min

Course content

  1. AI Risks

    Adversarial robustness, ML supply chain, dual-use, threat scenarios.

  2. Attacks

    FGSM / PGD, data poisoning, backdoors, automated LLM jailbreaks.

  3. Transparency & Explainability

    Interpreting decisions, auditing tools, flipped classroom in groups.

Indicative schedule

Day 1 — Foundations & Attacks

  1. 09:00–09:45 Introduction — AI Risks Landscape lecture
  2. 09:45–10:30 Attacks Overview — Taxonomy & Threat Models lecture
  3. 10:30–10:45 Break
  4. 10:45–11:30 Adversarial Examples — FGSM & PGD lecture
  5. 11:30–12:15 Data Poisoning & Backdoor Attacks lecture
  6. 12:15–13:15 Lunch
  7. 13:15–14:45 Lab 1 — Attacking an Image Classifier lab
  8. 14:45–15:00 Break
  9. 15:00–15:45 LLM Security & Jailbreaking Techniques lecture
  10. 15:45–16:30 Threat Scenarios & Defensive Mindset lecture

Day 2 — Defenses, Explainability & Peer Teaching

  1. 09:00–09:45 Transparency & Explainability — Concepts lecture
  2. 09:45–10:30 Interpreting Model Decisions in Practice lecture
  3. 10:30–10:45 Break
  4. 10:45–12:15 Lab 2 — LLM Jailbreaking (black-box & white-box) lab
  5. 12:15–13:15 Lunch
  6. 13:15–14:00 Lab 3 — Explainability Tooling & Model Auditing lab
  7. 14:00–14:45 Flipped Classroom — Groups 1 & 2 flipped
  8. 14:45–15:00 Break
  9. 15:00–15:45 Flipped Classroom — Groups 3 & 4 flipped
  10. 15:45–16:30 Synthesis, Q&A & Closing Discussion lecture

Instructors

Pricing & logistics

  • CHF 3,800/person
  • CHF 24,000 team package (up to 8 participants)
  • Language: English or French
  • Location: on-site client or remote, lab environment provided

References

Organize this training

Adaptation possible to your internal stacks (in-house models, MLOps tooling) and your threat scenarios.

Organize this training

Request this training — AI Security — Attacks and Defenses

Tell us what your team needs. Reply within 24h.

Format