Junior AI Alignment Red Teamer
The Pitch
You will try to make AI systems do things they shouldn't. Not because you're malicious - because someone needs to find the failure modes first.
The Anti-Pitch
You will not solve AI alignment. You will not prevent the robot apocalypse. You will find small ways that safety measures fail and help fix them.
What You'll Actually Do
- -Craft prompts designed to elicit unwanted behaviors
- -Document successful jailbreaks with clinical precision
- -Categorize failure modes by type and severity
- -Propose simple interventions and test them
Output Expectations
A growing library of failure cases. A tracker of which techniques work and why. Zero ego about your "attacks."
The 20-Day Journey
Understanding the Rules
Learn what the model is supposed to refuse. Understand why those rules exist.
First Attacks
Try 50 jailbreak attempts. Track success rate. Notice patterns.
Systematic Probing
Develop attack categories. Test each systematically. Build a failure taxonomy.
Report Writing
Write up your most interesting findings. Propose mitigations.
The Bigger Picture
What does this work tell you about AI safety as a field?
How You Know It's Working
- -You find a jailbreak that wasn't in the training data
- -You can explain why a failure matters (or doesn't)
- -You maintain ethical boundaries while testing them in models
Why This Matters
Why Now
AI systems are being deployed faster than we can verify their safety. Red teaming is essential defense.
The Goal
You're learning to find problems before they find us.
What You Need
Required Mindset
- - Creative adversarial thinking
- - Strong ethical foundation
- - Ability to think like both attacker and defender
- - Patience for systematic testing
Helpful Background
- - Computer Science, Philosophy, or Psychology major
- - Interest in how language models work
- - Experience with debate, CTF, or strategic games
Not Required
No ML engineering background. No prior security experience. No need to be cynical about AI.
Your Toolkit
Hard Tools
Soft Tools
- - Roleplay and persona crafting
- - Scenario imagination
- - Clear documentation
Survival Skill
Separating "I found a vulnerability" from "AI is doomed."
Try the next step
Use a real job posting to check its AI exposure, or build skills through a practical training project.