“…For example, an ML model may be trained to predict privacy-sensitive information such as sexual orientation and/or political affiliation using social network profiles [11,21,27]. Recent work has shown similar intrusive usages of ML models, predicting attractiveness using profile pictures [9,13], and reporting life satisfaction [1,22]. Although some of these usages seem innocuous, the deployment of such ML models may be seen as a threat to individual privacy.…”