Data-driven learning analytics (LA) exploits artificial intelligence, data-mining, and emerging technologies, rapidly expanding the collection and uses of learner data. Considerations of potential harm and ethical implications have not kept pace, raising concerns about ethical and privacy issues (Holstein & Doroudi, 2019; Prinsloo & Slade, 2018). This empirical study contributes to a growing critical conversation on fairness, equity, and responsibility of LA lending mentor voices in the context of an online mentorship program through which undergraduate students mentored secondary school students. Specifically, this study responds to a phenomenon shared by four mentors who recounted hiding from mentees that they had seen their LA data. Interviews reveal the convergent and divergent ideas of mentors regarding LA in terms of: (1) affordances and constraints; (2) scope and boundaries; (3) ethical tensions and dilemmas; (4) paradoxical demands; and (5) what constitutes fairness, equity, and responsibility. The analysis integrates mentor voices with Slade and Prinsloo's (2013) principles for an ethical framework for LA, Hacking's (1982, 1986) dynamic nominalism, and Levinas's (1989) ethics of responsibility. Design recommendations derived from mentor insights are extended in a discussion of ethical relationality, troubling learners as data-subjects, and considering the possibilities of the agency, transparency, and choice in LA system design.