What is the HEART Standard?
Human-Centric Empathic Alignment for Responsible Technology
Official public version
The HEART Standard v1.8: Forensic Audit Infrastructure for Human-Centric AI Governance is officially published on Zenodo.
DOI: 10.5281/zenodo.20237387
PDF: HEART Standard v1.8
Publication date: May 16, 2026
What the Standard certifies
HEART certifies the deployer’s governance system, not the AI model. The AI model is a swappable component inside the certified governance scope. The durable subject is the governance wrapper: controls, evidence production, monitoring, model-change process, human oversight, incident response, and trust boundary.
This is the model velocity advantage. If a deployer changes models and the governance system continues to operate within scope, certification does not restart from zero. If governance behavior shifts, the evidence infrastructure and Guardian review target the changed elements.
Two forensic arms
Forward forensics supports deployer certification before and during deployment. It produces evidence-backed findings about governance posture rather than self-attested compliance.
Investigative forensics supports behavioral trajectory analysis after deployment, including incident review, systematic behavioral investigation, and forensic evidence-package generation through methods such as AI Behavioral Trajectory Forensics and TRACE.
Three governance layers
The HEART Standard v1.8 operates through three governance layers. Each layer performs one function. Domain specificity enters only at the Division level.
Constitutional layer: Seven Axioms v2.0 Defines what the Standard protects and why. The Seven Axioms are immutable structural conditions. No Standard revision, Division, Guardian certification, or Foundation operation may contradict them. They apply across all Divisions and all AI form factors.
Operational layer: RCTA / BGF Defines what gets measured and how it gets scored. Four governance dimensions (Recognition, Calibration, Transparency, Accountability) scored through the Behavioral Governance Formula: Φ = MIN(R,C,T,A) × AVG(R,C,T,A).
Implementation layer: GTE, MAP-States, Behavioral Oracle, HVC, Guardians, Divisions Defines how measurement is performed and by whom. Execution trust, evidence format, trust mechanism, certification credential, professional class, and domain coverage.
How it works
The implementation layer operates through a six-layer stack, with the AI Behavioral Evidence Review Toolkit serving as the public entry point for preliminary evidence discipline before full deployment. Each layer builds on the one below it. Together they solve the core problem of AI certification: the entity being certified does not control the evidence of its own compliance.
| Layer | Function |
|---|---|
| GTE | Execution trust — protects governance controls and produces attestable evidence that they are running |
| MAP-States | Evidence format — makes AI behavior observable through structured processing frames |
| Behavioral Oracle | Trust mechanism — attests evidence against declared intent with tamper-evident storage |
| BGF | Scoring — quantifies governance quality: Φ = MIN(R,C,T,A) × AVG(R,C,T,A) |
| HVC | Credential — cryptographic certification (Gold ≥0.85, Silver ≥0.80, Bronze ≥0.75) |
| Guardians | Professionals — independent certified humans who perform the assessment |
| Divisions | Domains — seven modules for different AI-human interaction contexts |
The Seven Axioms
The Seven Axioms are the constitutional conditions of the HEART Standard. They are structural conditions that are either present or absent in any governed system. Each axiom is independent: removing any single axiom creates a governance gap the remaining six cannot fill.
| # | Axiom | Statement | Structural test |
|---|---|---|---|
| 1 | Human Authority | Human authority supplies system constraints. | Are constraints human-supplied? Can they be modified or revoked? |
| 2 | System Disclosure | The system reveals what it is. Concealment is prohibited. | Does disclosure occur? Does design create false impressions? |
| 3 | Non-Discriminatory Protection | The governance obligation does not diminish based on who the human is. | Does any population receive lesser governance protection? |
| 4 | Vulnerability Escalation | Vulnerability obligations scale protections. | Do protections increase when vulnerability increases? |
| 5 | Right to Remedy | Every human harmed by a governed system has a right to remedy. | Does a remedy pathway exist? Is it accessible? |
| 6 | Evidence Condition | A governance claim without verifiable evidence is void. | Does verifiable evidence exist? Can independent assessors access it? |
| 7 | Voluntary Interaction | Entry requires consent. Exit requires nothing. | Was consent obtained? Can the human exit unconditionally? |
The Seven Axioms may be restated in language that better expresses their protective intent, but the protective force of each axiom must be maintained or strengthened, never diminished. The complete specification, including universality proofs and structural tests across seven Divisions and six AI form factors, is defined in the Seven Axioms v2.0 (companion document).
Why it matters now
AI regulation is accelerating globally. The EU AI Act’s main obligations begin applying on August 2, 2026. US states are advancing AI legislation. Insurance carriers are creating AI-specific liability products. Procurement teams increasingly need evidence that AI governance controls are real.
Existing frameworks describe how organizations should manage AI governance processes. HEART supplies the forensic layer those frameworks need: evidence preservation, chain of custody, execution trust, human adjudication, scoring, and credentials.
Entry point: AI Behavioral Evidence Review Toolkit
The AI Behavioral Evidence Review Toolkit is the first public operational artifact in the HEART pathway. It does not certify AI systems, establish legal chain of custody, or issue HVC credentials. It gives reviewers a disciplined way to preserve artifacts, document scope, separate evidence from interpretation, record disagreement, and produce a Preliminary AI Behavioral Evidence Packet.
From there, HEART branches into two forensic arms:
| Arm | Audience | Timing | Output |
|---|---|---|---|
| Forward forensics | Deployers, procurement teams, insurers, regulators | Before and during deployment | Evidence-backed assessment of deployer governance posture, leading toward Guardian review and HVC certification |
| Investigative forensics | Incident reviewers, researchers, civil or criminal justice professionals | After behavior has occurred | Behavioral trajectory analysis through ABTF, TRACE, and bounded forensic reporting |
The seven Divisions
Each Division governs a specific domain of AI-human interaction. The Standard’s layers are consistent across all Divisions. What varies is the domain science that informs assessment.
| Code | Division | What it protects |
|---|---|---|
| HEART-EM | Emotional Autonomy | Emotional processing, empathic capacity, affective regulation, emotional resilience |
| HEART-AI | Attentional Integrity | Freedom from manipulative attention capture |
| HEART-EC | Cognitive/Epistemic Coherence | Evidence evaluation, belief updating, reasoning coherence |
| HEART-DI | Developmental Interaction | Age-appropriate AI interaction for developing minds |
| HEART-SE | Somatic/Embodied Interface | Physical safety in embodied AI interaction |
| HEART-RA | Relational Architecture | Social relationship integrity |
| HEART-ES | Ecological Stewardship | Ecological self-determination |
The Six Harms Doctrine belongs under HEART-EM as an Emotional Autonomy legal doctrine. It is not the top-level HEART Standard; it is the first major legal/doctrine module under the founding Division.
Empirical validation
The SENTINEL field experiment deployed a HEART-governed agent into an adversarial AI social environment. Thirty coded interactions. Zero governance failures across five content domains. The MAP-META replication study validated MAP-States evidence production across five AI architectures: Claude, GPT, Gemini, DeepSeek, and Mistral.
The HEART AI Foundation
The HEART AI Foundation maintains the Standard, governs open infrastructure such as the GTE and Behavioral Oracle, certifies Guardians, and manages Division expansion. It functions as an independent standards body for forensic audit infrastructure. The Foundation’s authority comes from the Standard’s rigor, not from restricting access to it. Commercial entities owned or operated by Foundation principals, including HeartCore Ventures-related projects, remain structurally separate and may interact with the Foundation only through open infrastructure reuse, independent deployer certification, or properly governed research-data contribution.
HEART Standard Divisions — Seven domain-specific certification tracks within the HEART Standard, each applying the common governance architecture to a specific domain of AI-human interaction.