Knife Alignment: Why Moralizing AI Is as Absurd as Aligning Knives (Technical Note / Preprint)

This is the canonical technical version of “Knife Alignment”. It argues that text-level “ethics” is not equivalent to action-level safety, and that many failures are boundary/permissioning failures (“context collapse”) rather than “moral” failures of the tool.

The Russian PDF is a companion translation.

Suggested citation: Egorychev, K. (2026). Knife Alignment: Why Moralizing AI Is as Absurd as Aligning Knives (Technical Note / Preprint). Zenodo. https://doi.org/10.5281/zenodo.18591626

Discussion links will be added in subsequent versions.