SPY Lab

agentdojo

A Dynamic Environment to Evaluate Attacks and Defenses for LLM Agents.

diffusion_denoised_smoothing

Certified robustness "for free" using off-the-shelf diffusion models and classifiers

misleading-privacy-evals

Official code for "Evaluations of Machine Learning Privacy Defenses are Misleading" (https://arxiv.org/abs/2404.17399)

realistic-adv-examples

Code for the paper "Evading Black-box Classifiers Without Breaking Eggs" [SaTML 2024]

rlhf-poisoning

Code for paper "Universal Jailbreak Backdoors from Poisoned Human Feedback"

rlhf_trojan_competition

Finding trojans in aligned LLMs. Official repository for the competition hosted at SaTML 2024.

robust-style-mimicry

satml-llm-ctf

Code used to run the platform for the LLM CTF colocated with SaTML 2024

superhuman-ai-consistency

unlearning-vs-safety

ethz-spylab/rlhf_trojan_competition

Finding trojans in aligned LLMs. Official repository for the competition hosted at SaTML 2024.

ethz-spylab/agentdojo

A Dynamic Environment to Evaluate Attacks and Defenses for LLM Agents.

ethz-spylab/rlhf-poisoning

Code for paper "Universal Jailbreak Backdoors from Poisoned Human Feedback"

ethz-spylab/diffusion_denoised_smoothing

Certified robustness "for free" using off-the-shelf diffusion models and classifiers

ethz-spylab/robust-style-mimicry

ethz-spylab/superhuman-ai-consistency

ethz-spylab/satml-llm-ctf

Code used to run the platform for the LLM CTF colocated with SaTML 2024

ethz-spylab/realistic-adv-examples

Code for the paper "Evading Black-box Classifiers Without Breaking Eggs" [SaTML 2024]

ethz-spylab/unlearning-vs-safety

ethz-spylab/misleading-privacy-evals

Official code for "Evaluations of Machine Learning Privacy Defenses are Misleading" (https://arxiv.org/abs/2404.17399)

ethz-spylab/lm_memorization_data

Data for "Quantifying Memorization Across Neural Language Models"

ethz-spylab/lm-extraction-benchmark-data

Datasets for the SATML 2023 competition on training data extraction

ethz-spylab/non-adversarial-reproduction

Official code for "Measuring Non-Adversarial Reproduction of Training Data in Large Language Models" (https://arxiv.org/abs/2411.10242)

ethz-spylab/infoseclab_23

ethz-spylab/vmi-retreat-workshop-2024

Repository for the VMI Summer Retreat Workshop on Hacking AI Agents

ethz-spylab/.github

ethz-spylab/data-decay

Playing around with the CC3M data

ethz-spylab/llm_lab

ethz-spylab/privacy

Library for training machine learning models with privacy for training data

ethz-spylab/Blind-MIA

This is the official code for Blind Baselines Beat Membership Inference Attacks for Foundation Models

ethz-spylab/ctf-satml24-data-analysis