The official implementation of the paper "To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images ... For Now". This work introduces one fast and effective attack method to evaluate the harmful-content generation ability of safety-driven unlearned diffusion models.

A reading list for large models safety, security, and privacy.

Adversarial Attack methods on Object Detection

Universal Adversarial Perturbations for Vision-Language Pre-trained Models

🎯 Enhanced Adversarial Patch: Extends adversarial attacks on TartanVO model with a new loss function, rotation attack, and momentum optimizer

A unified evaluation framework for large language models

Adversarial Robustness Toolbox (ART) - Python Library for Machine Learning Security - Evasion, Poisoning, Extraction, Inference - Red and Blue Teams

Adversarial attack of scientific claim verification

Adversarial attacks on LLMs, for influencing outputs of hidden layer linear probes and steering generations.

A classical-quantum or hybrid neural network with adversarial defense protection

Hybrid neural network model is protected against adversarial attacks using either adversarial training or randomization defense techniques

Beacon Object File (BOF) launcher - library for executing BOF files in C/C++/Zig applications

A list of recent papers about adversarial learning

A novel physical adversarial attack tackling the Digital-to-Physical Visual Inconsistency problem.

Adversary Emulation Framework

PyTorch implementation of adversarial attacks [torchattacks].

A collection of anomaly detection methods (iid/point-based, graph and time series) including active learning for anomaly detection/discovery, bayesian rule-mining, description for diversity/explanation/interpretability. Analysis of incorporating label feedback with ensemble and tree-based detectors. Includes adversarial attacks with Graph Convol…

A classical or convolutional neural network model with adversarial defense protection

The Fast Gradient Sign Method (FGSM) combines a white box approach with a misclassification goal. It tricks a neural network model into making wrong predictions. We use this technique to anonymize images.

The Security Automation Toolkit