VicByte Academy
AI Red Teaming Training
A hands-on weekend intensive covering how AI systems actually get attacked: prompt injection, jailbreaks and data exfiltration, delivered with a specialist offensive security partner.
Why AI security skills matter now
AI adoption is outpacing AI security. The gap is where attackers operate.
A new attack surface
LLMs and autonomous agents are being wired into core business systems faster than they are being secured.
Real incidents, rising
Prompt injection, data leakage, model abuse and agent misbehaviour are now showing up in production. The latest OWASP Top 10 for LLM Applications is built on a review of more than 6,600 real incidents.
Governance pressure
Boards and regulators expect secure-by-design AI, aligned to frameworks such as the NIST AI RMF.
A skills gap
Very few practitioners can test AI systems with a genuine attacker mindset. This course builds that capability.
What you'll learn
Learn the attacks by performing them, then learn how to defend against them.
Prompt injection
How attackers manipulate model inputs to bypass intended behaviour, and how to test for it.
Jailbreaking techniques
Practical methods used to circumvent safety guardrails, and what defenders should watch for.
Data exfiltration
How sensitive data can leak through AI system outputs, and how to close those paths.
OWASP Top 10 for LLM Applications
- Prompt injection
- Sensitive information disclosure and system prompt leakage
- Excessive agency and improper output handling
- Supply chain risks, and data and model poisoning
- Unbounded consumption, misinformation, and vector and embedding weaknesses
MITRE-aligned AI attacks
- Advanced model manipulation and misuse
- Persistent agent abuse and autonomous exploitation
- Retrieval poisoning against RAG architectures
- Stealthy exfiltration pathways aligned to ATLAS
Offensive techniques
- Crafting context-aware jailbreaks
- RAG data exfiltration
- Bypassing prompt guardrails and filters
- Function-calling abuse and code execution
Red teaming methodology
- Structured test planning for LLM architectures
- AI-specific threat modelling and scoping
- Mapping multi-stage attack chains
- Actionable reports with remediation steps
Category names follow the OWASP Top 10 for LLM Applications, published by the OWASP GenAI Security Project under CC BY-SA 4.0.
Curriculum
Six progressive modules, from fundamentals to a full red team exercise.
- 1
Foundations
AI and ML basics for security, risk models and threat taxonomy.
- 2
LLM and agent threats
A deep dive into the OWASP Top 10 for LLM Applications.
- 3
MITRE mapping
ATLAS and ATT&CK technique alignment and kill chains.
- 4
Offensive labs
Prompt injection, jailbreaks, data exfiltration through RAG and connectors, and agent abuse.
- 5
Detection and hardening
Monitoring, guardrails, policy and SDLC integration.
- 6
Capstone red team
A full red team exercise on a live lab application, with reporting and remediation.
Real-world lab experience
Practise against real AI systems, not slides.
- Private sandbox
- LLM applications
- Agent frameworks
- RAG stacks
- API tool integrations
- 24/7 lab access
- Live scoring and leaderboards
Private lab environment
A fully provisioned private sandbox with complex LLM applications, agent frameworks, RAG stacks and modern API tool integrations, with guided attack chains.
Challenge-based scenarios
Realistic, multi-stage challenges that simulate enterprise environments and emerging AI attack surfaces, drawn from more than five years of live AI penetration testing and red teaming engagements.
Attack and defence loops
Exploit a weakness, analyse the detection, apply hardening and re-test, in repeated cycles, with practical remediation playbooks on hand.
Trainer and teaching assistants
A lead trainer is joined by dedicated teaching assistants, so questions are answered while you work through the labs.
24/7 lab access
The lab is available around the clock, so you can keep practising outside session hours.
Safe by design
Isolated sandboxes with deep backend logging, transparent telemetry, real-time objective scoring and leaderboards.
Who this is for
Pentesters and red teamers
Extend traditional skills to AI attack surfaces and build modern attack chains.
AppSec and product security
Secure LLM features, build robust guardrails and shift AI security left.
SOC and threat hunters
Adapt detections, telemetry and response playbooks to autonomous AI systems.
AI, ML and MLOps teams
Learn the attacker view to build resilient, secure-by-design models and RAG systems.
Security consultants
Upskill to deliver AI security assessment and advisory work to clients.
What you take away
Measurable skills
Hands-on labs produce quantifiable results through challenge scores, attack artifacts and performance telemetry.
Portfolio-ready reports
Produce professional reports and exploit scripts from the lab exercises to demonstrate real capability.
Certificate and digital badge
An official certificate of completion and a verifiable digital badge to validate your AI offensive security skills.
Deployable playbooks
Leave with templates, operational playbooks and methodologies you can apply to enterprise AI red teaming.
Register your interest
Dates and pricing are coming soon. Leave your email and you'll hear first when the course opens.