AI agents & tool execution
Scoped authority, attempt identity, replay protection, STOP and receipts.
Where the control pattern can be applied.
Applications are use patterns beneath capabilities, not renamed internal programmes or unsupported product claims.
Machine purchasing & transactions
Purchase authority, budget boundaries, uncertain payment handling and delivery/clearing evidence.
Institutional & healthcare handoffs
Explicit release, bounded projection, receiver admission, responsibility-transfer truth and reconciliation.
Compute & data-centre workloads
Workload admission, distributed execution, STOP and evidence with SIL/real-GPU ceilings kept explicit.
Critical-infrastructure information
Provenance, canonical state and bounded advisory/API consumption; no implied physical actuation.
Autonomous physical systems
Contextual authority, bounded action, STOP and recovery; current proof remains simulator/software-bounded.
Cross-domain cooperation
Bounded projections, semantic interchange, independent admission and retained local authority.
Human-governed automation
Analysis/proposal, escalation, exact human decision, fresh revalidation and execution.
Disconnected / mission systems
Local authority, degraded communications, finite-resource operation, STOP and reconciliation; software/SIL ceiling unless stronger evidence exists.