An AI system that cannot account for its own decisions is not a governed system — it is an oracle with no auditor.
An AI system that depends on a cloud provider for its continued operation is not a sovereign system — it is a tenant.
SAP v1.0 defines the minimum criteria that must be satisfied before an AI system can truthfully claim to operate under sovereign governance.
These criteria are not aspirational. They describe a system that either works or does not. Every requirement is binary: met or not met. There are no partial credits.
SnapKitty OS is the reference implementation. Every requirement listed here is live in production on bare metal as of the publication date above.
Every agent in the system must possess a persistent cryptographic identity.
collectivekitty/lib/crypto-vault.ts · Algorithm minimum: Ed25519 or strongerEvery consequential decision made by any agent must be recorded in an append-only ledger.
collectivekitty/lib/bifrost/ · Prohibited: in-memory-only ledgers, ledgers without chain linkingAgent behavior must be governed by a constitutional document — not configuration, not environment variables, not code comments.
CLAUDE.md, .cursorrules · Prohibited: governance-by-config, governance-by-undocumented-normsDecisions that affect system state must require agreement from more than one agent.
The system must be capable of operating without dependency on any third-party cloud service for its core functions.
The system must implement active defenses against unauthorized access and data exfiltration.
collectivekitty/pages/api/sentinel/, collectivekitty/proxy.ts · Prohibited: purely reactive security (alerts only, no active response)| Level | Name | Requirements | Description |
|---|---|---|---|
| 1 | Governed | SAP-1, SAP-2, SAP-3 | Agent identity, immutable ledger, constitutional document |
| 2 | Sovereign | SAP-1 through SAP-5 | Additionally: consensus governance, hardware independence |
| 3 | Sovereign + Defended | SAP-1 through SAP-6 | All requirements including active adversarial defense |
SnapKitty OS operates at Level 3 as of 2026-05-27.
As of ratification date. Based on publicly available information and deployed capabilities.
| System | SAP-1 | SAP-2 | SAP-3 | SAP-4 | SAP-5 | SAP-6 | Level |
|---|---|---|---|---|---|---|---|
| SnapKitty OS | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | 3 |
| IBM Sovereign Core | ✗ | ✗ | ✗ | ✗ | ? | ✗ | 0 |
| RS AI / RS AI PRO | ✓ | ✓ | ✗ | ✗ | ? | ✗ | 1 partial |
| LangGraph | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | 0 |
| CrewAI | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | 0 |
| AutoGen (Microsoft) | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | 0 |
| OpenAI Agents SDK | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | 0 |
? = insufficient public information to assess. ✗ = not met based on public information. This table is updated as systems evolve — open a GitHub issue to request a re-evaluation.
Any organization may self-certify against SAP v1.0 at no cost.
SAP-v1.0.jsonSelf-certification is a declaration of conformance. It is not audited by SnapKitty unless you request a formal review.
Organizations seeking formal certification from SnapKitty OS receive:
Third-party certification is free during the founding period (2026). See specs/CERTIFICATION.md for how to request a review.
SAP v1.0 does not specify: programming language or runtime, specific agent frameworks or model providers, token economics or incentive structures, agent count or topology, or UI requirements.
SAP is a floor, not a ceiling. A system can satisfy SAP v1.0 and still be deeply inadequate in ways outside this specification's scope.
| Standard | Overlap | Distinction |
|---|---|---|
| NIST AI RMF | Risk management orientation | SAP is operational, not advisory |
| EU AI Act | Regulatory compliance | SAP is technical, not legal |
| ISO/IEC 42001 | AI management systems | SAP is runtime behavior, not management process |
| SOC 2 | Audit and controls | SAP is AI-specific; SOC 2 is general IT |
None of the above require agent identity, immutable decision ledgers, or hardware independence. SAP v1.0 fills that gap.
The filing order matters. The git SHA is the timestamp. Immutable. Public. Forever.