AttaX-Multimodal: End-to-End Evaluation of Multimodal AI Safety with Threatscore and Resiliencescore

Abstract

Currently, there is no single benchmark that can be used to measure the safety and resilience of multimodal AI assistants when subjected to malicious attacks. To fill this gap, we have created AttaX-Multimodal, a comprehensive benchmark of multimodal AI assistant safety that tests text, image, audio, and video-based AI assistants, as well as cross-modal attack chains and tool use attack methods. The contribution of this work includes a large-scale dataset of multimodal adversarial examples and a

Categories

Framework mappings

Suggested from the entry's categories.

@inproceedings{karne2026attaxmultimodal,
  title = {{AttaX-Multimodal: End-to-End Evaluation of Multimodal AI Safety with Threatscore and Resiliencescore}},
  author = {Rahul Karne},
  year = {2026},
  month = apr,
  booktitle = {2026 International Conference on Connected Intelligence for Industrial Applications (CI2A)},
  doi = {10.1109/CI2A69097.2026.11577059},
  url = {https://www.semanticscholar.org/paper/62fef5fdd0b48cd8d9acaab2d40663825eb1a98f}
}