EngineersofAI LogoCoursesBlogResearch LabAI LettersThe LabCode BankInteractive 3DKodrEarnestJobs
Skip to main content
Interactive 3D/Adversarial Prompts & Red Teaming
Attack Type
Defense
HUD
None
Bypass prob: 0%
Benign query - no attack.
Defense-in-depth:
Input validationOutput filteringRate limitingSemantic classifiers

Adversarial Prompts & Red Teaming - Interactive Visualization

Red teaming LLMs means systematically finding prompts that bypass safety guardrails. Common techniques include prompt injection (inserting instructions in retrieved content), role-play jailbreaks (asking the model to pretend it's uncensored), and token manipulation (using leetspeak or Unicode to evade classifiers). This demo shows each attack type and defensive countermeasures.

  • Prompt injection - see how malicious instructions embedded in retrieved content hijack the model's behavior
  • Role-play jailbreaks - visualize how persona-framing attempts to suppress the model's safety training
  • Token manipulation - see how replacing characters with Unicode lookalikes or leetspeak evades keyword classifiers
  • Defensive countermeasures - compare output filtering, input sanitization, and constitutional critique against each attack
  • Red team scoring - understand how automated red teaming pipelines rate attack success and iterate on prompts

Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.