Trust-Domain Externalisation Pattern.

Three-domain separation for independent verification of AI execution authority. 20 claims.

An architectural pattern that separates a sovereign AI runtime into three independently-controlled trust domains: an action-proposing domain (where the brain or agent emits a candidate action), a perimeter-enforcing domain (where typed-action gates, biometric checks, and policy enforcement run), and an audit-witness domain (where the signed ledger is appended). Each domain runs in a separate OS process, container, or hardware boundary so the perimeter cannot be silently bypassed by a compromise of the action-proposing layer. Filed 4 May 2026 as GB2610415.8.