Citation and attribution

We developed BEHA³VE for the benchmark track of the Workshop on Agent Behavior at COLM 2026, building on our ABxLab shopping experiments and interventions.

Related paperCopy Anchor Link

If you find BEHA³VE useful in your research, please cite our ABxLab paper:

Manuel Cherep, Chengtian Ma, Abigail Xu, Maya Shaked, Pattie Maes, and Nikhil Singh. A Framework for Studying AI Agent Behavior: Evidence from Consumer Choice Experiments. ICLR 2026.

@inproceedings{cherep2026abxlab,
  title={A Framework for Studying AI Agent Behavior: Evidence from Consumer Choice Experiments},
  author={Cherep, Manuel and Ma, Chengtian and Xu, Abigail and Shaked, Maya and Maes, Pattie and Singh, Nikhil},
  booktitle={The Fourteenth International Conference on Learning Representations},
  year={2026},
  url={https://arxiv.org/abs/2509.25609}
}

In this paper, we evaluated agents in shopping environments. BEHA³VE extends that work to additional environments and experimental designs.

Sources and license noticesCopy Anchor Link

We retain the shopping-source license in the ABxLab attribution notice.

We adapted local HTML pruning from BrowserGym and retain its Apache 2.0 notice. We also retain the OSWorld Apache 2.0 notice for our OSWorld integration.

These notices apply to the attributed components. We have not specified a package-wide license for BEHA³VE. Dependencies retain their own license terms.

We use Libre Franklin and IBM Plex Mono through Fontsource and retain both font licenses: Libre Franklin and IBM Plex Mono.