Advanced · Responsible AI & Security
NEWA4AI Security & Red Teaming
Stay ahead of attackers: find, test, and fix the security weaknesses unique to AI and LLM systems.
Who it's for
Security engineers, developers, and solution architects who build or secure AI and LLM-based applications.
You'll walk away with
A red-team assessment report and mitigation plan for an LLM application
Identify, test, and mitigate security threats specific to machine learning and LLM-based systems, including AI agents.
Format
- Lectures + hands-on security labs
Assessment
Pre- and post-tests; red-team lab and report
Tools & platforms
- garakPyRITGuard modelsLLM guardrail frameworks
Programme Curriculum
What you'll be able to do
- Explain the AI threat landscape and the most common vulnerabilities of LLM applications
- Conduct structured red-team testing, including prompt injection and jailbreak evaluation
- Implement defenses: input/output filtering, guardrails, and least-privilege tool access
- Integrate AI security into the software development lifecycle and governance
Your learning journey
Module 1
The AI Threat Landscape
- Threats & Attack SurfacesCommon LLM application vulnerabilities; data, model, application, and agent attack surfaces
Module 2
Attacking AI Systems
- LLM & Agent ThreatsDirect and indirect prompt injection; jailbreaks; excessive agency; data leakage in RAG
- Model & Data ThreatsData poisoning; model extraction; supply-chain risks
Module 3
Defending AI Systems
- Defense in DepthGuardrails and guard models; sandboxing agents; secrets and access control; monitoring and incident response
Module 4
Red-Team Lab
- Structured Red TeamingTest planning; automated scanning with open-source tools; report and remediation plan
Remarks
- Prerequisite: Basic programming and application security knowledge; familiarity with LLMs