aisafety

[NeurIPS 2024 Oral] Aligner: Efficient Alignment by Learning to Correct

Toolkit for research purposes in AIS. See the website for the paper.

[CoRL'23] Adversarial Training for Safe End-to-End Driving

AI-Generated Video Detection via Perceptual Straightening (NeurIPS2025)

[ACL 2025 Findings] Fraud-R1 : A Multi-Round Benchmark for Assessing the Robustness of LLM Against Augmented Fraud and Phishing Inducements

[ACL 2025 Findings] The official GitHub repo for the paper "Nuclear Deployed: Analyzing Catastrophic Risks in Decision-making of Autonomous LLM Agents"

AI defense infrastructure against manipulation, misuse, hallucinations, and synthetic deception.

Can Large Language Models Solve Security Challenges? We test LLMs' ability to interact and break out of shell environments using the OverTheWire wargames environment, showing the models' surprising ability to do action-oriented cyberexploits in shell environments

A high-performance string formatter written in Rust. This project detects and blocks LLM prompt injection and jailbreak attacks. It also features a customizable rule-based system and defends against obfuscated prompt attacks.

An official repository for the Capability-Based Scaling Laws for LLM Red-Teaming paper.

[NeurIPS 2024] SACPO (Stepwise Alignment for Constrained Policy Optimization)

A benchmark for evaluating hallucinations in large visual language models

Multi-agent simulation using LLMs. Agents autonomously decide actions for survival, reproduction, and social behavior in a grid world.This project aims to replicate a paper published in 2025 (arXiv:2508.12920).

Learned Semantic Decoder for Language Models.- Its the little model that sits under a big model's hat to explain what its thinking, just like the little cat's from Cat in the Hat! VOOM > FOOM

The official github repo for [LinguaSafe paper](https://arxiv.org/abs/2508.12733)

Safe Option Critic: Learning Safe Options in the A2OC Architecture

AI Safety Evaluation Library

Explore techniques to use small models as jailbreaking judges

Three-layer deterministic safety & ethical execution guard for autonomous robots (ROS 2 & Behavior Trees). Sub-ms physical reflex (QERRA-HSR) + moral filter (SEMEV-12) + values ranker (THRIVE). Fully auditable.

This open-source blueprint provides a structured, rigorous methodology for creating credible, accessible, and actionable whitepapers across any domain—especially technical, research, AI safety, strategic, or high-stakes fields.