Fictional role simulation. This is a learning scenario created by 10xCareer, not a live opening or an offer from a named employer.
The Adversarial Friend

Junior AI Alignment Red Teamer

AI Safety|San Francisco, CA or Remote|Anthropic - Red Team

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

1
Days 1-3

Understanding the Rules

Learn what the model is supposed to refuse. Understand why those rules exist.

2
Days 4-6

First Attacks

Try 50 jailbreak attempts. Track success rate. Notice patterns.

3
Days 7-12

Systematic Probing

Develop attack categories. Test each systematically. Build a failure taxonomy.

4
Days 13-18

Report Writing

Write up your most interesting findings. Propose mitigations.

5
Days 19-20

The Bigger Picture

What does this work tell you about AI safety as a field?

How You Know It's Working

Level: Undergraduate
  • -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

API access to modelsSpreadsheets for trackingBasic scripting

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.

← Back to Sandbox