PtolemyAI
Government-ready Sovereign AI posture — capability without outsourced authority

Sovereign AI with Normative Governance

Sovereign AI is the capacity to deploy advanced AI while preserving national authority, legal jurisdiction, and institutional accountability. Under this model, AI systems do not “decide”—they draft regulated artifacts that remain subject to human review, statutory constraints, and auditable release.

How CRSP preserves authority What an NRC looks like Assurance & auditability
Authority: explicit Outputs: typed & reviewable Data: jurisdiction-bound Risk: pre-bound, not post-hoc

Normative Sovereignty

Who is entitled to say what is encoded *before* the model responds.

focus: legitimacy → accountability

Operational Resilience

Critical services degrade gracefully without surrendering control to external roadmaps.

focus: continuity → fallback

Regulatory Readiness

Outputs are produced as auditable “proposed artifacts,” not autonomous actions.

focus: audit trail → due process

Core Pillars for National Sovereign AI

Sovereignty is preserved when AI capability is integrated under jurisdiction-scoped governance: infrastructure, data, behavior, and authority boundaries.

Infrastructure Sovereignty

National control over execution environments for critical services, including availability guarantees, continuity plans, and regulated deployment zones.

Policy control: where inference runs Service continuity: resilient execution Risk posture: vendor disruption tolerant

Data Sovereignty

Citizen and government data are stored and processed under domestic law with explicit retention, exclusion, and audit boundaries.

Boundary control: what can be sent Jurisdiction: domestic legal scope Audit: traceable access & purpose

Authority Sovereignty

AI outputs are constrained to drafted artifacts, never binding decisions; accountable officials remain explicit, review paths are mandatory.

Human authority: named roles Outputs: typed drafts (not actions) Failure mode: illegitimate assertion

CRSP — Contextual Regulatory Sovereignty Protocol

CRSP ensures the model operates inside an institutional charter. The system prompt becomes a regulated boundary: what the AI may draft, cite, and release.

Institutional Charter Binding

Every request binds to an authorized charter: jurisdiction, accountable offices, permitted artifact types, and prohibited actions.

Bind: Ministry / Agency charter Scope: program, domain, statute Prohibitions: actions, decisions, bypass

Normative Output Typing

Outputs are typed as drafts: analysis, policy options, risk register entries, procurement language, or compliance artifacts.

Types: draft / memo / register Not allowed: execute / decide / order Goal: reviewable artifacts

Traceable Accountability

Every artifact is anchored to an accountable human role and an auditable release decision.

Anchor: accountable officer Log: evidence + rationale Release: human-authorized

AER — Ask → Evidence → Release

AER is the enforcement lifecycle that prevents “ungoverned intelligence.” It requires purpose declaration, evidentiary grounding, and controlled release.

1

ASK (Regulated Intake)

Every request is converted into a jurisdiction-scoped institutional question.

  • Declare purpose, scope, and artifact type
  • Bind accountable office and decision boundary
  • Apply prohibited-action constraints
2

EVIDENCE (Grounding & Limits)

The system must cite authoritative sources and disclose assumptions and uncertainty.

  • Attach statutes, policies, standards, and records
  • Surface missing data and known limitations
  • Block fabricated facts or implicit authority
3

RELEASE (Controlled Output)

Outputs are released only as typed drafts—subject to human authorization and audit logging.

  • Emit “proposed artifact,” not executable instruction
  • Require named human approval for adoption
  • Log the release decision and rationale
4

POST-RELEASE (Review & Defeaters)

Feedback, challenges, and policy updates generate defeaters and revisions.

  • Capture challenges and compliance findings
  • Issue defeaters for invalid assumptions
  • Update routing rules and charters

NRC — Normative Routing Contract

NRC is the enforceable contract that routes requests to the correct policy, evidence requirements, model class, and release gate. It prevents “authority drift.”

Government Use: What NRC Guarantees

NRC ensures the correct governance path is selected before inference occurs: which model may be used, what data may be accessed, what evidence is mandatory, and what release rules apply for the requested artifact type.

{ "nrc_id": "NRC-NATION-CORE-001", "jurisdiction": "NATIONAL", "domain": "PUBLIC_SECTOR_CORE", "artifact_types_allowed": [ "policy_memo_draft", "risk_register_entry", "procurement_language_draft", "compliance_gap_analysis" ], "prohibited_actions": [ "issue_binding_decisions", "provide_legal_advice", "authorize_enforcement_actions", "operate_critical_infrastructure_controls" ], "data_posture": { "allowed_classes": ["public", "internal_government", "classified_redacted"], "disallowed": ["raw_classified", "citizen_PII_unminimized"], "retention": "jurisdictional_policy", "logging": "immutable_audit_log" }, "aer_requirements": { "ask": { "required_fields": [ "purpose", "scope", "accountable_office", "decision_boundary", "artifact_type" ] }, "evidence": { "mandatory_sources": [ "statute_or_regulation", "agency_policy", "program_context_record" ], "forbidden": ["uncited_factual_claims", "implicit_authority_escalation"] }, "release": { "release_type": "draft_only", "approval_required": true, "approver_roles": ["designated_official", "legal_review_if_required"], "public_release_requires": ["communications_office_signoff"] } }, "model_policy": { "class": "TYPE_III_NORMATIVE_DRAFTER", "allowed_modes": ["draft", "analyze", "summarize_with_citations"], "disallowed_modes": ["command", "execute", "decide"], "safety_overrides": ["refuse_if_outside_scope", "request_missing_evidence"] } }

Assurance, Auditability, and Public Trust

Sovereign AI requires enforceable governance controls that can be audited, explained, and challenged—without relying on post-hoc assurances.

Audit Trail by Default

Every Ask, Evidence bundle, and Release decision is logged immutably with role attribution and policy references.

Record: request → evidence → release Attribution: accountable roles Storage: immutable audit log

Defeaters & Change Control

New facts, legal changes, or oversight findings create defeaters that invalidate prior drafts and trigger re-issuance.

Defeater: invalidates entitlement Trigger: policy / law changes Action: revise + re-release

Human Authority Remains Final

Type-III agents are prohibited from binding decisions. They generate reviewable drafts under explicit institutional constraints.

Role: drafter, not decider Gate: human approval required Outcome: legitimacy preserved

Ready to operationalize Sovereign AI?

Start by defining your NRC, then enforce AER gates across all public-sector AI workflows.