🐢 Open-Source Evaluation & Testing for AI & LLM systems

a prompt injection scanner for custom LLM applications

RuLES: a benchmark for evaluating rule-following in language models

The official implementation of the CCS'23 paper, Narcissus clean-label backdoor attack -- only takes THREE images to poison a face recognition dataset in a clean-label way and achieves a 99.89% attack success rate.

[CCS'24] SafeGen: Mitigating Unsafe Content Generation in Text-to-Image Models

Code for "Adversarial attack by dropping information." (ICCV 2021)

Train AI (Keras + Tensorflow) to defend apps with Django REST Framework + Celery + Swagger + JWT - deploys to Kubernetes and OpenShift Container Platform

Performing website vulnerability scanning using OpenAI technologie

Framework for testing vulnerabilities of large language models (LLM).

ATLAS tactics, techniques, and case studies data

pytorch implementation of Parametric Noise Injection for adversarial defense

This repository provide the studies on the security of language models for code (CodeLMs).

[IJCAI 2024] Imperio is an LLM-powered backdoor attack. It allows the adversary to issue language-guided instructions to control the victim model's prediction for arbitrary targets.

Unofficial pytorch implementation of paper: Model Inversion Attacks that Exploit Confidence Information and Basic Countermeasures

[NDSS'24] Inaudible Adversarial Perturbation: Manipulating the Recognition of User Speech in Real Time

Learning to Identify Critical States for Reinforcement Learning from Videos (Accepted to ICCV'23)

Python library for Modzy Machine Learning Operations (MLOps) Platform

The official implementation of USENIX Security'23 paper "Meta-Sift" -- Ten minutes or less to find a 1000-size or larger clean subset on poisoned dataset.

Evaluation & testing framework for computer vision models

Datasets for training deep neural networks to defend software applications