This companion volume extends the Ethical AI Series — Conservation of Intent into the adjacent jurisdictions of alignment and safety while preserving the trilogy's sealed posture, lawful labels, and terminal refusal boundary. It treats alignment and safety language as classification-only objects governed by authorized representations, bounded domains, and binding non-claims rather than as mechanisms, mandates, interfaces, enforcement regimes, or outcome guarantees. The manuscript defines a companion ontology for alignment-preserving and safety-preserving interpretation, specifies admissibility and conformance conditions under institutional and architectural pressure, and preserves refusal wherever lawful classification would require operational substitution, prohibited inference, outcome capture, or silent scope expansion. Throughout, the work remains non-operational, non-inferential, and non-certifying: it does not authorize identity inference, mental-state inference, behavioral inference, surveillance posture, evaluation mandates, deployment-readiness claims, or assurances that a system is aligned or safe. Its purpose is narrower and more durable: to preserve lawful meaning, constrain derivative misuse, and keep alignment and safety language citable without becoming capturable.
Matthew A. Davis (Sun,) studied this question.