The Criterion of Loss: Why AI Cannot Yet Be a Moral Subject

This essay argues that artificial intelligence systems cannot yet be treated as moral subjects, and that the decisive criterion for moral subjecthood is the capacity for irreversible loss — that is, the capacity for a being to lose something of its own in a way that matters to it. Drawing on this criterion, the essay distinguishes between functional agency and moral presence, critiques the widespread conflation of scale with threshold in AI discourse, and identifies the human withdrawal from responsibility as the central danger in the deployment of AI systems. The concept of Hashomer (the guardian) is introduced as an architectural and ethical function that preserves the structure of human accountability in the age of powerful AI.