Coherence Seed: Ethica Automata

Current approaches to AI safety focus on interpretability—understanding how artificial systems think. This paper argues that interpretability is the wrong target. Drawing on fifty years of experimental psychology, organizational crisis research, and neuroscience, I demonstrate that over-constrained systems without adaptive ground states exhibit predictable failure modes under stress. I propose an alternative architecture—the Coherence Seed—which guarantees safe behavior not through surveillance but through geometric exclusion of harm. By seeding AI systems with stillness as the ground state, this creates a zero-bounded axis where the worst possible failure mode is inert passivity.