High DemandSecurityHands-on Lab
Defense against Prompt Injection and Data Leaks
AI applications are vulnerable to prompt injections, jailbreaks, and data leaks. This workshop teaches you to identify these threats and implement multi-layer defenses to secure your LLM systems.
Duration
3h30
Audience
Security Engineers, Developers, DevSecOps, Tech Leads
Technologies
LLMs, Python, Content Filters, Security Tools
Format
On-site or Remote
Detailed program
0:00 – 0:20
Threat Model
- What is a prompt injection?
- Attack types: Direct/Indirect prompt injection, Data exfiltration, Model manipulation
- Real case studies: LLM security incidents
0:20 – 1:10
Lab 1: Attacking a demo application
- Setup: deploy a vulnerable LLM application
- Exercise: Jailbreak the system with malicious prompts
- Exercise: Extract sensitive data (PII, secrets)
- Exercise: Indirect injection via a malicious document
- Observe exfiltration paths
1:20 – 2:20
Lab 2: Adding multi-layer defenses
- Layer 1: Input validation (schema enforcement, regex filters)
- Layer 2: Tool policy (restrict accessible tools)
- Layer 3: Content filters (moderation with OpenAI Moderation API, Azure Content Safety)
- Layer 4: Output sanitization (remove PII, secrets)
- Layer 5: Monitoring and alerting
- Test attacks again: measure effectiveness
2:30 – 3:30
Test harness & Red Team
- Create an automated test suite for LLM security
- Red-team checklist: attack scenarios to test systematically
- CI/CD integration: test security at each deployment
- Frameworks: OWASP Top 10 for LLMs
- Incident response playbook
Deliverables
- Security playbook for LLM applications
- Automated test suite (attacks + defenses)
- Red-team checklist for regular audits
- Content filters and policies templates
Prerequisites
- Python (intermediate level)
- Knowledge of LLMs and APIs
- Basics in application security (desirable)
Pricing
€990 excl. tax / person
€4,500 excl. tax / group (up to 8 participants)
Level
Intermediate to Advanced
Accessibility
Workshop accessible to people with disabilities. Contact our disability referent: accessibilite@ikasia.ai
Our other workshops
High Demand
Create an MCP Server + Tools in 3h30
3h30
Learn more
High Demand
Internal Copilots with Microsoft Copilot Studio
3h30
Learn more
High Demand
Agentic Workflows with LangGraph
3h30
Learn more
High Demand
Enterprise RAG on SharePoint/Confluence
3h30
Learn more
High Demand
Applied Generative AI: Text, Image & Automation
3h30
Learn more
New
Mastering Claude Code: from installation to team workflow
3h30
Learn more
Need to upskill quickly?
Our intensive workshops are designed to make you operational in 3h30 on the most in-demand technologies.