[Feature Request] Red Team / Security Testing Support

[Leo-Lok]

Hi team,
EVA currently has adversarial tests for business process bypass (e.g., skipping verification, unauthorized data access).

Would be great to also support security red team testing — e.g., prompt injection, jailbreak attempts, voice-specific social engineering.

[fanny-riols]

Absolutely! That's something on our roadmap, but we currently don't have the bandwidth. However, we're open to external contributions. If you would like to extend our existing datasets, or add new ones specifically for security testing, please go ahead! Here is a blog post we wrote explaining how we created data: https://huggingface.co/blog/ServiceNow-AI/eva-bench-data