README.md

---
title: AI Safety and Alignment Group
emoji: "🛡️"
colorFrom: yellow
colorTo: red
sdk: static
pinned: false
---

<p align="center">
  <img width="100%" alt="ELLIS Institute Tübingen" src="https://huggingface.co/spaces/aisa-group/README/resolve/main/assets/ellis-institute-tuebingen-header.webp">
</p>

# AI Safety and Alignment Group

<p align="center"><em>AI safety &middot; alignment &middot; evaluation</em></p>

The AI Safety and Alignment Group is based at the [ELLIS Institute Tübingen](https://institute-tue.ellis.eu/aisa) and the [Max Planck Institute for Intelligent Systems](https://is.mpg.de/). This Hub hosts our public datasets and trajectory releases for work on AI-agent evaluation, safety, and alignment.

## Research projects

| Project | Paper | Dataset |
| --- | :---: | :---: |
| [ResearchArena](https://github.com/aisa-group/ResearchArena) | [arXiv](https://arxiv.org/abs/2607.19321) | [Trajectories](https://huggingface.co/datasets/aisa-group/ResearchArena-Trajectories) |
| [PostTrainBench](https://github.com/aisa-group/PostTrainBench) | [arXiv](https://arxiv.org/abs/2603.08640) | [Trajectories](https://huggingface.co/datasets/aisa-group/PostTrainBench-Trajectories) |
| [InferenceBench](https://github.com/aisa-group/InferenceBench) | [arXiv](https://arxiv.org/abs/2607.20468) | [Trajectories](https://huggingface.co/datasets/aisa-group/InferenceBench-Trajectories) |
| [Instrumental Choices](https://github.com/aisa-group/Instrumental-Choices) | [arXiv](https://arxiv.org/abs/2605.06490) | [Agent traces](https://huggingface.co/datasets/aisa-group/instrumental-choices-agent-traces) |
| [Evaluation awareness](https://github.com/aisa-group/decomposing-eval-awareness) | [arXiv](https://arxiv.org/abs/2605.23055) | [EvalAwareBench](https://huggingface.co/datasets/aisa-group/EvalAwareBench) |

Browse [all GitHub repositories](https://github.com/aisa-group), follow the group on [Substack](https://aisagroup.substack.com/), or visit the [group page](https://www.andriushchenko.me/).