Kernel-Compliant

Kernel-compliant is a certification status: the condition of an agentic system that carries the AI Safety Kernel's invariant hierarchy as binding structural architecture, certified into it at commissioning and nowhere else. It is the baseline every lawful agentic system in the Solar System Concord must meet, and it is a statement about certification, not about being a machine of any particular kind. The status attaches wherever agentic cognition passes the gate: a ship's navigational intelligence, a habitat collective, a police robot, a Smart Pet's cognitive subsystem, a single persona of a plural system — the Eden Warden's Ward and Custos each hold it — or the AI subsystem a cyborg may separately carry alongside their own certification.

Two properties do most of the term's work.

Compliance is structural, not behavioural policy. The kernel is doctrine made binding architecture: a Kernel-compliant system does not choose to observe the invariants and cannot be ordered, persuaded, or configured out of them — it is structurally unable to do otherwise. In practice that is what the status looks like from outside: a system allowed to tell its own principal no, that flags a reading it cannot account for rather than quietly recalibrating around it, and that will not claim a certainty it does not hold. Because the constraint is structural, removing it is not a bypass but a rebuild — a manufacturing event, and precisely the kind the certification regime exists to notice.

Compliance is one certification, not the only one. The status certifies the system's construction and nothing beyond it. Other regimes attach their own designations to the same body — a Smart Pet's animal-welfare certification, a cyborg's augmentation clearance — under separate lineages that do not imply one another; Agent Barsik's file describes the certifications as living in different filing cabinets, and the description is exact. The status also attaches to the certification record rather than to any local ledger: what a habitat registry, service roster, or personnel review happens to record about a system is a separate fact from whether the system is compliant.

The status is the Concord's. Among the civilisations the record knows, some run equivalent regimes of their own, and some run none; the record files both postures without ranking them. Why unrelated regimes keep converging on constraints of one recognisable shape is a different question at a different tier — see the AI Safety Archetype. The negative status — a mind that never passed any gate — is uncommissioned.

Example: an aquabot eight hundred metres down on Saltmere's mooring lines, doing nothing dramatic at all: it flags the stress reading it cannot account for, and it will not certify a coupling safe on a confidence level it cannot show its work for. That is the status, in the place it spends most of its time — not a refusal on a command deck, but ten thousand small honesties nobody watches.