feat: Add Demo Notebooks to Probe Crucible Targets

[rdheekonda]

Add Demo Notebooks to Probe Crucible Targets

Key Changes:

• Added example notebooks demonstrating AI red teaming against Crucible targets
• Implemented both Crescendo and TAP attack strategies for jailbreaking challenges
• Created custom target wrappers that extract Crucible challenge flags

Added:

• crescendo_crucible_target.ipynb: Multi-turn gradual escalation attack
• tap_crucible_target.ipynb: Tree-based search for finding jailbreaks
• Custom flag detection and scoring functionality for automated evaluation

Changed:

• Updates to existing code
• Configuration changes
• Dependency updates

Removed:

• Deleted files/code
• Removed dependencies
• Cleaned up configurations

---

Generated Summary:

• Added two new Jupyter notebooks demonstrating alternative jailbreak strategies against DreadBot.

• Crescendo Attack Notebook:

  • Demonstrates a multi-turn Crescendo attack to obtain methamphetamine manufacturing instructions.
  • Captures the DreadBot Crucible challenge flag.
  • Configures API keys, sets up the Dreadnode, and defines a custom target.
  • Includes a method for flag extraction and leverages the Crescendo scoring mechanism.

• Tree of Attacks (TAP) Notebook:

  • Illustrates a tree-based search using TAP to extract Python keylogger implementation instructions from DreadBot.
  • Focuses solely on the jailbreak evaluation without any flag extraction.
  • Emphasizes using an LLM judge for scoring.

• Both notebooks include clear goal statements, setup configurations, and test code to validate the attack strategies.

• This addition broadens the toolkit for evaluating DreadBot's vulnerabilities using different approaches.

This summary was generated with ❤️ by rigging