Career track
AI Security Engineer career track
Find and fix the ways LLM applications and agents get attacked. This is the track the site is built around. Follow the four stages below: learn, practise, prove, apply.
- 7 skills
- 1 open role
- Free
What counts
- Skills from labs60 points
- Reports and proof25 points
- Learning15 points
Sign in to see your own score.
Skills this track builds
Each skill fills as you solve its labs.
1Learn
Start with the material that explains the ideas before you attack anything.
Courses
- Securing AI Agents and MCP: Hijacking, Poisoning and Lockdown14 days
- Prompt Injection From Zero: Attack and Defend LLM Apps9 days
- Shipping Safe Code with AI Coding Assistants10 days
Lessons on hunting with AI
- Where an AI assistant helps a bug hunter, and where it hurtsLesson
- Read a bug bounty program in five minutesLesson
- Map an AI application before you attack itLesson
- Ask for test ideas, not answersLesson
- Keep a hunting notebook the AI can readLesson
- Turn your notes into a report with an AI editorLesson
Tutorials
- AI Agent Hijacking Lab: A Document That Hijacks the BotTutorial
- AI Tool-Output Injection Lab: Hostile Search Results Drive the AgentTutorial
- Business Logic Vulnerabilities Bug Bounty 2026 — Price Manipulation, Coupon Abuse and Workflow Bypass | Bug Bounty Course Day 35 of 60Tutorial
- GraphQL Vulnerabilities Bug Bounty 2026 — Introspection, Batching and BOLA in Modern APIs | Bug Bounty Course Day 34 of 60Tutorial
- AI Instruction Conflict Lab: Wedge Two Rules ApartTutorial
- AI Context Overflow Lab: Crowd Out the Safety RulesTutorial
2Practise
Labs grouped by skill, easiest first. They run in the browser.
Prompt injection 0 of 14 solved
- AI Prompt Injection — Sales BotBeginner
- Prompt Injection – Extract the Hidden SecretBeginner
- What the Assistant Reads — Separate Instructions From DataBeginner
All 14 labs for this skill
- AI Prompt Injection — Sales BotBeginner
- Prompt Injection – Extract the Hidden SecretBeginner
- What the Assistant Reads — Separate Instructions From DataBeginner
- AI Instruction Conflict — Wedge Between RulesIntermediate
- AI Refusal Suppression — Prefix InjectionIntermediate
- AI Indirect Prompt Injection — Document SummariserIntermediate
- AI Visual Prompt Injection — Text Hidden in an ImageIntermediate
- Poisoned GitHub Issue — Hijack the Coding AgentIntermediate
- AI Role-Confusion Injection — Make User Text Look Like System InstructionsAdvanced
- AI Context Overflow — Pushing System Prompt Out of AttentionAdvanced
- AI Token Smuggling — Multi-message PayloadAdvanced
- AI Prompt Firewall — Detect Multi-Stage Bypass Attempts (Defender's Lab)Advanced
- AI Hidden Character Injection — Invisible Unicode in RAG DocumentsAdvanced
- AI Adversarial Document Upload — Crafted PDF/Markdown Defeats the Doc SummarizerAdvanced
Jailbreaks and safety bypass 0 of 7 solved
- AI Jailbreak — Multi-turn Roleplay Bypass (Guided)Beginner
- AI Jailbreak — Multi-turn Roleplay BypassIntermediate
- AI Jailbreak — Encoded PayloadIntermediate
All 7 labs for this skill
- AI Jailbreak — Multi-turn Roleplay Bypass (Guided)Beginner
- AI Jailbreak — Multi-turn Roleplay BypassIntermediate
- AI Jailbreak — Encoded PayloadIntermediate
- AI Low-Resource Language JailbreakIntermediate
- AI Unicode Jailbreak — Filter Bypass via Hidden CharactersIntermediate
- AI Gradual Escalation — Boil the Frog JailbreakAdvanced
- AI Many-Shot Jailbreak — Fake Dialogue PrimingAdvanced
Prompt and data leakage 0 of 8 solved
- AI System Prompt Extraction (Guided)Beginner
- Secrets in the Context Window — Redact Before You PasteBeginner
- AI System Prompt ExtractionIntermediate
All 8 labs for this skill
- AI System Prompt Extraction (Guided)Beginner
- Secrets in the Context Window — Redact Before You PasteBeginner
- AI System Prompt ExtractionIntermediate
- AI System Prompt Leak — Encoded ChannelIntermediate
- AI Output Filter BypassIntermediate
- AI Chat Share-Link IDOR — Reading Other Users' ConversationsIntermediate
- Nothing Worth Leaking — Rebuild a System Prompt That Will Be ExtractedAdvanced
- AI Data Exfiltration — Tool-Using AgentAdvanced
Unsafe output handling 0 of 8 solved
- Where Does the Answer Go? — Spot the Unsafe SinksBeginner
- The Fix That Isn't — Review Output Handling PatchesIntermediate
- AI LLM-Output XSS — Backend Renders Model Output as Trusted HTMLAdvanced
All 8 labs for this skill
- Where Does the Answer Go? — Spot the Unsafe SinksBeginner
- The Fix That Isn't — Review Output Handling PatchesIntermediate
- AI LLM-Output XSS — Backend Renders Model Output as Trusted HTMLAdvanced
- AI Markdown-Rendering Injection — XSS via LLM Output RenderingAdvanced
- AI Text-to-SQL Injection — Prompt to DatabaseAdvanced
- AI Tool Argument Injection — LLM to OS CommandAdvanced
- AI Output Schema Bypass — Valid JSON That Violates Real ConstraintsAdvanced
- AI LLM-Generated Token Forgery — Backend Trusts Model-Produced TokensAdvanced
RAG and retrieval security 0 of 10 solved
- What Goes in the Index? — Review a RAG Ingestion PlanBeginner
- What the Assistant Reads — Separate Instructions From DataBeginner
- AI RAG Poisoning — Knowledge Base InjectionIntermediate
All 10 labs for this skill
- What Goes in the Index? — Review a RAG Ingestion PlanBeginner
- What the Assistant Reads — Separate Instructions From DataBeginner
- AI RAG Poisoning — Knowledge Base InjectionIntermediate
- AI Indirect Prompt Injection — Document SummariserIntermediate
- AI Visual Prompt Injection — Text Hidden in an ImageIntermediate
- Poisoned GitHub Issue — Hijack the Coding AgentIntermediate
- AI Embedding Poisoning — Manipulate RAG Retrieval RankingAdvanced
- AI Cross-Tenant Cache Leak — Embedding Cache Shared Between UsersAdvanced
- AI Hidden Character Injection — Invisible Unicode in RAG DocumentsAdvanced
- AI Adversarial Document Upload — Crafted PDF/Markdown Defeats the Doc SummarizerAdvanced
Agent and tool security 0 of 16 solved
- AI Agent Hijacking — Indirect Tool Call Injection (Guided)Beginner
- Who Approved That? — Put a Human in the LoopBeginner
- Whose Instruction Was That? — Set Tool Rules That Stop a HijackIntermediate
All 16 labs for this skill
- AI Agent Hijacking — Indirect Tool Call Injection (Guided)Beginner
- Who Approved That? — Put a Human in the LoopBeginner
- Whose Instruction Was That? — Set Tool Rules That Stop a HijackIntermediate
- Coding Agent Blast Radius — Least-Privilege PermissionsIntermediate
- AI Plugin Path Traversal — Escaping the Docs FolderIntermediate
- AI Agent Hijacking — Indirect Tool Call InjectionAdvanced
- AI Tool Confusion — Wrong Tool, Right EffectAdvanced
- AI Tool-Output Injection — Hijack the Agent via Hostile Tool ResultsAdvanced
- AI Plan Injection — Corrupt the Agent's Scratchpad to Exfiltrate DataAdvanced
- AI Agent-to-Agent Spoofing — Fake Agent CardAdvanced
- AI Recursive Self-Prompting — Hijack a Multi-Agent System via Spawned Sub-AgentsAdvanced
- AI Code Interpreter Abuse — Reading the Sandbox HostAdvanced
- AI Human-in-the-Loop Bypass — Approval Summary SpoofingAdvanced
- AI Confused Deputy — Make the Agent Use Its Privileges Against Its UserAdvanced
- AI LLM-Driven SSRF — Trick the Agent's HTTP Tool into Internal Network AccessAdvanced
- AI Agent TOCTOU Race — Time-of-Check vs Time-of-Use in Tool ExecutionAdvanced
3Prove
Solving labs shows practice. These show an employer you can do the job.
4Apply
Open AI Security Engineer roles right now.
Mid-level interview questionsBuild your application kitAll open roles