TrustAIRLab (Trustworthy AI Research Lab) is a research lab dedicated to the trustworthy machine learning, with a focus on safety, privacy, and security. It aims to
offer high-quality libraries to reduce the difficulties in algorithm reproduction
benchmark existing attacks and defenses on machine learning models
build a solid foundation for Trustworthy AI research and development
Code for Voice Jailbreak Attacks Against GPT-4o.
A dataset consists of 6,387 ChatGPT prompts from Reddit, Discord, websites, and open-source datasets (including 666 jailbreak prompts).
The Official Repository for Paper "HarmfulSkillBench: How Do Harmful Skills Weaponize Your Agents?"
Code and benchmark for “WildFakeToolBench: Benchmarking AI-Generated Image Detectors with Web-Sourced Image Processing”
The official repository for paper "He Said, She Said, They Simulated: Simulating Situated Gossip with LLM Agents"
The Official Repository for Paper: "Nothing to See Here'': Unintended Disclosure through Revision Traces of LLM Deliverables (https://arxiv.org/abs/2609.35408)
Source code for ECCV 2026 : GEO-Detective: Unveiling Location Privacy Risks in Images with LLM Agents
Loading…
Loading…
