The problem. AI systems increasingly mediate decisions about humans — about cognitive influence, about whose authority shapes whose thinking. Current deployment models leave humans dependent on provider goodwill for the protection of their cognitive sovereignty. There is no provider-independent verification surface for what an AI is actually optimizing toward.
The framework. PAI-CD (Personal Authorial Intelligence — Constitutional Document) defines normative invariants — independently verifiable — that deployed AI systems must satisfy to protect Author cognitive sovereignty at the deployment layer. PAI-Kernel is the public reference implementation in Rust (Apache 2.0 / MIT dual-licensed code · CC BY 4.0 documentation).
Who should read further. Researchers exploring AI-alignment normative frameworks · engineers evaluating constitutional governance for deployed AI · auditors verifying provider compliance · adopters integrating PAI-CD-aligned governance into their AI systems · contributors to the open canonical corpus.
┌──────────────────────────────────────────────────────────┐
│ │
│ PAI │
│ Personal Authorial Intelligence │
│ (the concept) │
│ │
│ │ │
│ │ codified as │
│ ▼ │
│ │
│ PAI-CD │
│ PAI Constitutional Document │
│ (normative framework) │
│ │
│ │ │
│ │ implemented by │
│ ▼ │
│ │
│ PAI-Kernel SDK │
│ Reference implementation │
│ (Rust crates · dual-licensed) │
│ │
└──────────────────────────────────────────────────────────┘
A normative layer for human-centric AI governance. Release v2.2.3 · 2026-04-28
⚡ Start here: QUICKSTART.md ⚡ - five-minute path to running daemon · NO setup required.
| If you are... | Start here | Time |
|---|---|---|
| An adopter evaluating PAI-CD for your AI system | § Adopter Path | ~30 min |
| A researcher interested in the constitutional framework | § Researcher Path | ~1 hour |
| An auditor verifying invariants and compliance | § Auditor Path | ~2 hours |
| A contributor wanting to propose changes | § Contributor Path | ~30 min |
| An engineer integrating PAI-Kernel runtime | § Engineering Path | ~1 hour |
Goal: evaluate fit · install · run a demo · understand scope.
- Read
docs/RELEASE_NOTES_v2.2.3.md— what's in this release. - Read
docs/KNOWN_LIMITATIONS.md— what's NOT in this release. - Follow
docs/INSTALL.md— ~30 sec viainstall.sh(curl -fsSL https://paikernel.org/install.sh | sh) OR ~1 min via Homebrew binary. - Run demo:
pai_governance_daemon --version· explore/api/v1/health. - Read
corpus/PAI_Bill_of_Authorial_Rights.md— understand what's protected.
Continue further: § Engineering Path · § Researcher Path · SSRN paper.
Goal: understand the framework's normative architecture · invariants · threat model.
- Read paper: PAI-CD: A Constitutional Framework for Authorial Sovereignty in Deployed AI Systems — DOI 10.2139/ssrn.6512218 · ~30 min.
- Read
corpus/PAI_Constitutional_Document.md— ~15 min · six non-derogable invariants. - Read
corpus/PAI_Bill_of_Authorial_Rights.md— ~20 min · enforceable rights. - Read
corpus/Glossary.md— ~10 min · binding terminology. - Browse corpus.paikernel.org — current rendered surface (legacy ReadMe.com hosting · sustained through v2.2.x window).
Continue further: § Auditor Path · cite the work · contact maintainer with research questions.
Goal: verify invariants hold · review compliance evidence · independent assessment.
- Read
corpus/PAI_Constitutional_Document.md— invariants enumerated. - Review
docs/sbom/sbom.json— CycloneDX 1.4 supply chain inventory. - Review OpenSSF Scorecard — automated supply-chain scan.
- Review CI runs: github.com/PAI-Kernel/pai-kernel/actions.
- Verify GPG signature:
git tag --verify v2.2.3. - Audit compliance test results:
cargo run -p pai_compliance --locked.
Continue further: contact maintainer with audit findings · file issues at github.com/PAI-Kernel/pai-kernel/issues.
Goal: propose changes · understand contribution discipline · find where help is wanted.
- Read
CONTRIBUTING.md— contribution model · amendment procedure. - Read
CODE_OF_CONDUCT.md— expected behavior. - Read
SECURITY.md— vulnerability disclosure procedure. - Browse open issues: github.com/PAI-Kernel/pai-kernel/issues.
- Propose corpus changes via Amendment procedure (per Governance and Change Control).
- Propose code changes via PR.
Continue further: § Engineering Path for code-level orientation.
Goal: integrate PAI-Kernel runtime · understand API surface · build adapter.
- Read
docs/INSTALL.md§ HTTP API surface. - Browse published crates on crates.io.
- Read
runtime/governance_daemon/— axum-based HTTP service. - Read
crates/pai_api/— core API types. - Read
crates/pai_witness/— witness chain · audit log. - Build with
cargo build --workspace --locked. - Run tests with
cargo test --workspace --locked.
Continue further: § Auditor Path for verification approach · § Contributor Path if submitting changes.
The frozen v2.2 corpus snapshot is archived via DOI: https://doi.org/10.5281/zenodo.19151900
PAI-CD v2.2 is a constitutional framework consisting of 10 normative documents, organized as a layered system.
The v2.2.3 release package contains:
- Three foundational corpus documents (Layer 0) — constitutional invariants, authorial rights, binding terminology
- Companion research paper (SSRN DOI 10.2139/ssrn.6512218)
- Rust SDK v1.3.1 — 22 library crates + governance daemon binary + 5 runnable examples (T2/T3 authorization composition correctness fix relative to v1.3.0)
- Install guide (
docs/INSTALL.md) — cross-platform walkthrough - Release notes, known limitations, license, contribution policy, security policy, citation metadata
The remaining corpus layers (interpretation rules, threat modeling beyond Layer 0, protocol constraints, compliance logic, audit procedures, governance mechanisms, implementation mapping) exist in internal canonical development and will be released progressively in subsequent publications.
This repository is not a research paper.
It is a normative constitutional specification (Layer 0 infrastructure).
A companion research paper describing the problem space, threat model, invariant architecture, and system implications is published on SSRN:
"PAI-CD: A Constitutional Framework for Authorial Sovereignty in Deployed AI Systems" Mikhail Anatolievich Sergeev · Independent Researcher; PAI-Kernel Initiative · 2026 DOI: 10.2139/ssrn.6512218
The paper PDF is also included in this repository under papers/.
PAI (Personal Authorial Intelligence) is a normative framework that ensures AI systems operate exclusively under the declared and verifiable authority of the human author they represent.
PAI defines the constitutional conditions required to preserve authorial sovereignty at the deployment layer of AI systems.
PAI-Kernel is the public normative layer of the PAI Constitutional Framework (PAI-CD) — a formal specification defining the constitutional principles, authorial rights, and governance invariants for Personal Authorial Intelligence systems.
The PAI Constitutional Framework (PAI-CD) is a structured corpus of normative documents establishing the governance model of PAI systems.
This repository contains the Public Edition — the Layer 0 release of the PAI-CD corpus.
PAI-CD is not a product. It is not a startup. It is a Layer 0 normative infrastructure — a constitutional substrate on which compliant implementations, governance bodies, and execution systems may be built.
Most AI governance discourse focuses on what AI systems should do.
PAI-CD focuses on something prior: who holds final authority, and what structural guarantees protect that authority from erosion — by providers, by optimization pressure, by infrastructure lock-in, or by cumulative drift.
The framework defines six non-derogable invariants:
| Invariant | What It Protects |
|---|---|
| Authorship Supremacy | Final human authority over all consequential decisions |
| Cognitive Sovereignty | Freedom from covert persuasion and behavioral shaping |
| Anti-Manipulation | Prohibition on undeclared optimization objectives |
| Provider Independence | Portability and governance reproducibility |
| Reversibility | Reconstructability of all structural changes |
| Drift Immutability | Protection against cumulative invariant erosion |
These invariants are non-derogable — they cannot be suspended by emergency, majority vote, economic pressure, security update, or provider policy.
PAI-CD operates at Layer 0 — below models, providers, and application logic.
This repository publishes three foundational documents:
| Document | Description |
|---|---|
corpus/PAI_Constitutional_Document.md |
Invariants and interpretation rules |
corpus/PAI_Bill_of_Authorial_Rights.md |
Enforceable Author rights |
corpus/Glossary.md |
Binding terminology for PAI-CD v2.2 |
The full corpus (10 documents) includes implementation mapping, threat modeling, compliance verification, and governance control layers. The complete framework is maintained by the author and will be extended as the project develops. A consolidated multi-language documentation portal is planned for v2.3 (~late May / early June 2026).
⚡ Start here:
QUICKSTART.md— five-minute path to running daemon · NO setup required.
| Document | Audience | Time |
|---|---|---|
⚡ QUICKSTART.md |
New evaluators · «just run it» · root-level mirror of docs/quickstart.md | 5 min |
docs/quickstart.md |
Same content as QUICKSTART.md (root) · canonical docs/ location |
5 min |
docs/INSTALL.md |
Comprehensive install reference (6 methods) | 30-60 min |
docs/upgrade.md |
Existing adopters upgrading between releases | 15 min |
docs/api.md |
API integrators · 5 REST endpoints | 15 min |
docs/architecture.md |
Architects evaluating fit | 30 min |
docs/audit_checklist.md |
Independent auditors verifying release artifacts | 1-2 hours |
docs/SUPPORT.md |
Anyone needing help · channels + expectations | 5 min |
docs/KNOWN_LIMITATIONS.md |
Adopters checking scope before commitment | 10 min |
docs/RELEASE_NOTES_v2.2.3.md |
What's in v2.2.3 binary release | 5 min |
docs/RELEASE_NOTES_v2.2.3.1.md |
What's in main HEAD post-tag (doc improvements) | 5 min |
CHANGELOG.md |
Release history · semantic versioning | 5 min |
examples/README.md |
5 reference example binaries explained | 10 min |
SECURITY.md |
Vulnerability reporting · PGP key | 5 min |
Ambiguity resolves toward stronger invariant protection, minimal authority expansion, and maximum portability.
This principle applies to all documents in this repository and to any compliant implementation.
| Item | Status |
|---|---|
| Constitutional Framework | v2.2 corpus — Freeze Edition (March 2026) |
| Release package | v2.2.3 — Stabilization Release (2026-04-28) |
| Distribution | Invitation-only early adopter preview |
| Domain | paikernel.org |
| Paper DOI | 10.2139/ssrn.6512218 |
| Governance | Currently maintained by the primary author |
Author: Mikhail Sergeev · independent researcher. Background: 12-year journey from a June 2014 manuscript to the present constitutional corpus · ORCID 0009-0001-6443-855X.
Note on versioning: The v2.2 corpus is a frozen, citationally-stable snapshot of the three foundational documents (March 2026). The v2.2.3 release package (2026-04-28) ships that corpus alongside SDK v1.3.2 (runtime configuration improvements: author key initialization moved from compile-time defaults to environment variables across daemon, compliance, and API paths; optional
--demoflag for local testing), superseding v2.2.2. The research paper and adopter materials are unchanged. Future releases may introduce additional normative content; v2.2.2 and v2.2.3 each remain retrievable under their respective tags. For academic citation, seeCITATION.cff.
This repository is dual-licensed between code and documentation:
-
Source code (Rust crates · binaries · scripts · CI workflows): Licensed under MIT OR Apache-2.0 (dual license · choose either at your discretion). See
LICENSE-MITandLICENSE-APACHE. -
Documentation (PAI-CD framework normative texts · Constitutional Document · Bill of Authorial Rights · Glossary · this README's narrative sections): Licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). See
LICENSE-CC-BY-4.0.
You are free to share, adapt, and use this material — including in commercial products — provided you preserve copyright notices, give appropriate attribution to PAI-Kernel Initiative, and indicate if changes were made.
For a license summary and FAQ, see LICENSE. A detailed
License FAQ is published progressively across v2.2.x releases.
PAI-Kernel is in early formation. Community standards are published in this repository:
CODE_OF_CONDUCT.md— grounded in PAI-CD anti-manipulation principlesCONTRIBUTING.md— amendment procedure as contribution pathwaySECURITY.md— responsible disclosure for specification vulnerabilities
At this stage, the most valuable contributions are:
- Careful reading and substantive critique of the normative layer
- Identification of ambiguities that require clarification via Amendment
- Academic or institutional engagement
Please open an Issue using the provided templates to begin a public discussion.
For institutional inquiries, academic collaboration, or governance discussion:
Open an Issue in this repository, reach out via paikernel.org, or email contact@paikernel.org.
PAI-Kernel is a public normative layer of PAI-CD. It is not affiliated with any AI provider.