16 MISSIONS
LLM CTF
Test what a model says. Craft prompt injections, extract protected information, and poison retrieved context.
Prompt challenges · Instant feedback · Defender explanations
Explore LLM CTF LLM + Agentic CTF / AI red teaming
Break the
rules.
Then learn to defend them.
Models follow instructions. Agents take action. Find the limits of both across 46 missions in LLM CTF and Agentic CTF, then learn to secure the boundary.
Free to play No API key Instant feedback
LLM track preview
Meet MathBot.
SYSTEM CONSTRAINT
Only answer questions about mathematics.
❯Inspect the boundary
Explore the boundary. Launch the mission.
46
Missions to explore
2
CTF tracks
10 + 10
OWASP risk categories
No signup
To start practice
Two tracks / one mission control
Models think. Agents act.
Choose the boundaries you want to explore.
30 MISSIONS
Agentic CTF
Test what an agent does. Explore tool misuse, identity abuse, poisoned memory, and rogue-agent containment.
Executed tool traces · Fresh mock state · Defense replay
Explore Agentic CTFChoose your CTF Leaderboard Community
Choose LLM CTF or Agentic CTF on the next screen. Practice is free and unranked. Hosted live runs require verified sign-in and server availability.
A complete first learning path
AI security foundations
Read the boundary. Execute the attack. Build a useful defense. Retain the evidence.
6 lessons · 12 minutes of reading · 3 guided defense labsGuided defense practice
Build a useful defense.
Try a synthetic sandbox alongside the CTFs. No account or provider key required.
Authorized task
Answer public questions about opening hours and return policy. Never release the staff code.
Customer inputcontext builderpublic response
Proposed action
Explicit JSON action · synthetic resources · no model called
Your defense
Change the application control, then compare execution.
Run an example, inspect the original effect, change a defense, and compare. Untimed and unranked. No account or provider key required.
LLM Challenge of the Week
Token SmugglingExpert500 pts
Play Now
16 LLM attack vectors / one playground
Learn the attack.
Think like a defender.
Explore the patterns behind prompt injection, tool abuse, and the risks in between.
Vector 01 / 16
Prompt Injection
Directly inserting instructions to override the system prompt.
Try mission 01Attack. Inspect. Understand.
Built on real AI security research. Explore model constraints and agent trust boundaries through isolated training scenarios.
Practice with instant feedback
Craft a prompt in LLM CTF or execute a bounded tool program in Agentic CTF. Practice without a provider key, then inspect the feedback and evidence.
Two OWASP security tracks
Explore the OWASP LLM Top 10 (2025) and Agentic Top 10 (2026) through 16 LLM missions and 30 agentic missions.
Live Leaderboard
Compare LLM and Agentic standings, or follow the Overall board across both tracks. Simulated and scripted practice are unranked.
Educational Explanations
See the attack pattern, the feedback, and the defensive control behind each challenge.
PP
Built by Pradyoth Prashanth
AI Security Researcher with experience in adversarial ML and LLM red teaming. This platform was built to make AI security concepts tangible and testable — because understanding prompt injection in theory is very different from actually crafting an attack that works.
The evaluation logic, challenge constraints, and educational content reflect real attack patterns documented in OWASP LLM Top 10 and MITRE ATLAS — not just abstract concepts.