Result description
This tool enables to assess the privacy risk of an AI model in terms of how vulnerable it is to leaking sensitive information about its training set. It can run both membership and attribute inference attacks, normalizes and aggregates the attack success scores and outputs an overall privacy health score for the model.
Addressing target audiences and expressing needs
- Fellowship to advance my/our research
Beta customers and collaborators.
- Research and Technology Organisations
- Academia/ Universities
R&D, Technology and Innovation aspects
This technology has been partially integrated into IBM’s watsonx.governance product.

