Gated Memory proposes a new formation framework for conversational AI memory systems, introducing an admission gate and conditional enrichment stage to improve memory quality. This framework evaluates candidate facts against the full utterance context before extraction and grounds admitted facts in an entity scope taxonomy with privacy constraints.