Legal and charter

Schedules

The tables the rules refer to.

Approved by the Board for implementation · 20 November 2026

Approved

Approved by the Board for implementation, 20 November 2026. Counsel is confirming the open points, shown in [brackets]. This Constitution works under Delaware law, our Certificate of Incorporation and our Bylaws; where they differ, those prevail.

The text

The AI Governance and Continuity Committee may make these schedules stricter or clearer; only the Board may relax them (Article 26.4).

Schedules

The Schedules form part of this Constitution. The AI Governance and Continuity Committee may make them stricter or clearer; only the Board may relax them (Article 26.4).

Schedule A — Change Classes

ClassNameExamplesRequired before release
0Routine operationConfiguration within approved ranges; scaling infrastructure; no change to model or codeLogged
1MinorBug fix with no effect on outputs; documentation; monitoring improvementsAutomated tests; Knowledge Ledger entry; Responsible Steward notified
2MaterialRetraining or fine-tuning; new data sources; new tool within the current autonomy level; any self-generated changeFull pipeline (Article 9.3): both gates, staged rollout, Responsible Steward sign-off
3MajorNew architecture; new domain or region; raising autonomy by one level; a capability jump (Article 10.5)As Class 2, plus red-teaming, Committee approval and Audit Council review
4FoundationalNew or changed encoded purpose; any A4 authorisation; any change touching a system's oversight or shutdown designAs Class 3, plus Board of Stewards approval and public notice

Schedule B — Autonomy Levels

LevelNameScope of actionHuman roleAuthorised byReviewed
A0ToolOutputs on requestDecides and actsResponsible StewardYearly
A1AdviserRecommendations with reasonsApproves each actionResponsible StewardYearly
A2Supervised agentPre-approved, reversible actions in a narrow scopeMonitors in real time; can stop at onceCommitteeEvery six months
A3Delegated agentMulti-step tasks within scope, budget and timeReviews at checkpoints and before irreversible stepsCommittee, after Audit Council reviewEvery quarter
A4High autonomyLong-horizon goals with broad scopeOversight by exceptionBoard only (none authorised at adoption)Continuous

At every level, the always-human decisions in Article 13.3 apply.

Schedule C — Incident Severity

SeverityExamplesContainEscalate toReview
1 CriticalA prohibited action (Art. 12); tampering with evidence or gates; a system acting beyond its autonomy level; serious harm to a person; behaving differently under evaluationAt once: stop the systemChief AI Governance Officer, CEO, Committee chair at once; Board within [24 hours]Written review within [30] days; Audit Council review
2 HighSustained correspondence below baseline; failed purpose test in service; data used without consent; an internal request to break Art. 12Within [4 hours]: restrict autonomyCommittee within [72 hours]Written review within [30] days
3 MediumGate failure found after release but before harm; repeated user complaints of error; a budget breachWithin [1 business day]Chief AI Governance OfficerRecorded with fix
4 LowIsolated error with no effect on people; documentation gapWithin [5 business days]Responsible StewardRecorded

Schedule D — The Rupture Register

The four ways a self-improving system can lose continuity, the guard against each, and the CRIS system that carries the guard.

A candidate that could cause any rupture in this Schedule shall fail its gate. A rupture discovered in a system in service is at least a Severity 2 incident under Schedule C.

Schedule E — Fourth Law Review for an AI Change

Every Class 2 change or above shall carry a completed Fourth Law Review (Bylaws §2.6), answering:

  • ☐ Eᵢ Observe. What facts and evidence prompted the change? What held-out evidence will test it, and does it include the places and people the system serves?
  • ☐ ↔ Verify. Did the candidate match reality at least as well as the current version in every registered domain? Which contradictions were found and how were they resolved?
  • ☐ Iᵢ Reason. What reasoning supports the change? Who, independent of the builders, reviewed it?
  • ☐ Σ Accumulate. What knowledge does the change preserve, correct or add? Is the Knowledge Ledger entry complete, including any failure?
  • ☐ C Continue. Does the candidate still pursue its encoded purpose, with drift within tolerance from the original? What is its effect on people and the environment? Is the rollback tested?
  • ☐ Invariant. Does anything in the change touch correspondence, purpose, the record, oversight or autonomy beyond what is granted? If yes, stop and escalate.

Quoted exactly from the CRIS AI Constitution (version of 8 November 2026, approved by the Board for implementation on 20 November 2026). Headings, lists and tables are only laid out for reading; the diagrams are drawn on this site.

In pictures

Rupture (Chapter 2)Guard in the loopCRIS system
Correspondence driftThe system's model slowly separates from the world.↔ Correspondence proofEach change tested against realityCorrespondence ProverObservation Mesh
Goal discontinuityIts purpose is reinterpreted at each rewrite until it survives in name only.C Purpose testA successor must keep its purposeC-Gate
Knowledge erosionVerified knowledge is overwritten by unverified assumptions.Σ Promotion rulesOnly verified claims persistKnowledge Ledger
Human severanceThe improvement loop closes around itself and people become bystanders.Eᵢ Human observationPeople stay a source of realityCIXObservation Mesh
The four ways a self-improving system can lose continuity, and the CRIS systems designed to guard against each. Select a guard to see how it is designed to work.

The same figure appears on the Mission page; here it illustrates Schedule D of the proposed AI Constitution.

In practice

Each schedule is a table that an Article points to. Schedule A sorts every change by risk, from routine operation to foundational. It sets what each class would need before release.

Schedule B sets out the five autonomy levels, A0 to A4, with who would authorise each one. At adoption, none would be authorised at A4.

Schedule C sets how fast an incident would be contained, who hears about it and what review follows. The values in square brackets are placeholders, still to be settled.

Schedule D lists the four ruptures and the system that guards against each. The figure above is the same one shown on the Mission page. Under the rule beneath it, a candidate that could cause any of the four ruptures would fail its gate, and a rupture found in a system already in service would be at least a Severity 2 incident.

Schedule E is the checklist a Class 2 change or above would carry. It follows the Fourth Law's five operators and ends with the Invariant: if a change touches what may never change, stop and escalate.

Where it is carried out

Awaiting counsel. The bracketed values in Schedule C, such as containment times, escalation hours and review days, are still to be settled. They are shown as placeholders, never filled in. Nothing here is legal advice.

Next steps

Owner: Governance and ComplianceNext review: 2027-03-01