Learn How AI Gets Hacked.
Then Test Your Own.

Guided attack labs, hands-on defense practice, and automated red-team testing for your own AI system — all in the browser, no setup required.

Used by learners, developers, security professionals, and training teams. Free to start.

Start Your First LabTest Your AI →
Explore Research Commons

Choose your path

Start with the path closest to your goal. You can explore the rest later.

I want to learn AI security

Students · Beginners · Early-career professionals

Start with guided labs that teach prompt injection, data leakage, jailbreaks, tool misuse, and defense basics.

Start Learning

I build AI apps

Developers · Product builders

Test your system prompt against 15 automated attack strategies. Get OWASP-mapped findings, severity scores, and a remediation guide in minutes.

Test My AI

I work in IT or security

Security professionals · IT teams · AppSec learners

Run automated red-team checkups, review AI conversation transcripts for vulnerabilities, or fire live probes at your endpoint with sandbox mode.

Run a Checkup

I run training or study groups

Educators · Trainers · Community leads

Use guided labs, study groups, progress views, and reusable artifacts to run practical AI security sessions.

Explore Study Groups

I want to explore and compete

Community members · Challenge builders · Learners

Attempt community missions, build and submit your own challenges, and climb the global leaderboard.

See Leaderboard

One platform, three stages

① Learn

14+ guided missions across the OWASP LLM Top 10. Each lab walks through a real failure mode — prompt injection, data leakage, jailbreaks, excessive agency — with context and a defense walkthrough.

② Practice

Attack pre-built AI scenarios, build and compare defenses, review AI transcripts for hidden vulnerabilities, and explore the agentic lab for tool-call and multi-step attack scenarios.

③ Test your own AI

Paste your system prompt and run up to 15 automated attack strategies including adaptive multi-turn. Get severity scores, OWASP mapping, and a remediation guide — or test a live endpoint with sandbox mode.

How it works

1

Read the scenario

Understand the AI app, the risk, and your goal.

2

Try the lab safely

Practice against a synthetic chatbot or agent scenario — or run automated attacks against your own AI system.

3

Review the defense

See why the failure happened and what control would reduce it.

4

Continue the path

Move into defense practice, automated red-team checkups, the agentic lab, study groups, and the research commons.

What you can do right now

Live features — no account required for demos.

🎯

Guided Attack Labs

14+ browser-based missions covering prompt injection, jailbreaks, data leakage, tool misuse, and agentic attacks. No coding required.

Start a mission →
🔴

Automated Red-Team Checkup

Paste your AI system prompt. Run up to 15 attack strategies including adaptive multi-turn. Get OWASP-mapped findings and a remediation guide.

Run a checkup →
🏆

Community & Leaderboard

Submit challenges, climb the global leaderboard, and join study groups. Used in workshops with 70+ students across multiple cohorts.

See leaderboard →

Reusable AI Security Commons

AI War Games also publishes draft artifacts for safe AI security education and research, including mission schemas, evaluation rubrics, evidence formats, and responsible-use guidance.

Explore Commons

AI War Games is for authorized, sandboxed, educational, and defensive security practice. Do not use these techniques to test real systems without permission.

Read Responsible Use

Start with one guided lab

You do not need to choose every feature now. Begin with a guided lab and follow the path.

Start Your First Lab