Jonathan Hung (Vanderbilt University) has posted Models as Personal Data on SSRN. Here is the abstract:
Machine learning models present inherent data privacy risks. While privacy attacks on models are still in their infancy, they are a rapidly growing threat in today’s artificial intelligence boom. Contemporary data privacy laws were not meant to address these risks. Even so, regulators can leverage these laws to mitigate the privacy risks within machine learning models by treating risky models as personal data. Doing so, however, creates a conflict between privacy and intellectual property interests. This Article attempts to resolve this conflict in a way which protects individual privacy interests while respecting intellectual property rights.