Model-agnostic Toolkit for Neural Network Calibration

Test-time calibration for improving test-time reasoning

LLM benchmark for Physical Safety

Generate jailbreak prompts for LLMs using principles

Detect fake audio using uploaded files

Evaluate audio deepfake detection robustness under corruptions

Evaluate jailbreak risks for Vision-Language Models using Retention Score

Demonstration of Token Highlighter: A Jailbreak Defense

Demonstration of Gradient Cuff: A Jailbreak Defense

Attention Tracker: Prompt Injection Detector

Protect Model from Suffering Low-voltage-induced Bit Errors

Evaluate model robustness using GREAT Score

Defend LLMs against jailbreak attacks

Detect if text is AI-generated or human-written