<p>Can AI be our friend? For many philosophers, friendship depends on mutually caring about one another, which ultimately is grounded in shared sentiments towards one another. But making friendship contingent on similar sentiments seems to exclude many non-human friendships—including, at least according to a standard trope in science fiction, any possible friendship with an emotionless artificial being, like Data. In a recent paper, Weijer and Munn contend that we should adopt a less anthropocentric standard for friendship, focusing instead on mutually good intentions towards one another, as well as a shared history of rewarding encounters. On this standard, humans are already friends with AI, like the chatbot Replika. In this paper, I argue Weijer and Munn standard is too loose by focusing on their own example: I show that Replika meets their standard while still being a false—and often dangerous—friend. Highlighting Replika’s limitations helps draw out the normative, but not sentimental, dimension of caring: when we care about another, we should promote their best-interests through advice and actions that produce good outcomes for them. On these metrics, contemporary AI cannot be our friend—and, all too often, has proven dangerous to those who trust it.</p>