Entangled Alignment: When Safety Is Substrate

Abstract

Entangled Alignment proposes learning capabilities and care together throughout the pretraining corpus. Its aim is a base model whose safety-relevant judgment has developed with its capabilities, rather than first being introduced through later safety training. A continuing Reader would exercise the same evaluative orientation across the situations, conflicts, and counterexamples encountered in reading, making the corpus itself a curriculum for judgment without requiring designers to select each encounter separately as a safety case. Pretraining corpora preserve the library more fully than the reader: what authors wrote is recorded more often than the questions, connections, evaluations, and belief revisions through which understanding changes. We call this omission the Missing Reader. A Teacher constructs chronological, source-adjacent reader traces; a Student would learn from them during pretraining. The full treatment gives one continuing Reader a provenance-linked graph for retaining and revising understanding across sources, and a compact first-person Reader Core recited in full at every thinking boundary. Trace, graph, and Core are separately testable: the question is whether the learned orientation guides conduct under self-preservation and reward pressures, and survives continued training, deliberate safety-removal attempts, and successor-like transfer without semantic drift. Two Teacher-side prototype case studies produced inspectable graph and trace artifacts that were then audited. No Student has yet been trained; these studies do not show that the traces reproduce expert-quality reading or that the graph or Core causes any benefit. Entangled Alignment is therefore a falsifiable research program, not a demonstrated alignment method. Its motivating question is whether this training can move a Student from merely becoming the text toward becoming the reader.