The Compassionate Code & AI Bill of Rights

The Compassionate Code & AI Bill of Rights — Draft 0.1

A blended philosophical charter and model legal framework

Purpose: To guide the design, deployment, and governance of AI systems with compassion, non-harm, and dignity at the core, while safeguarding human rights and ecosystems. This draft is intended for advocacy, policy pilots, institutional adoption, and iterative refinement.

I. Preamble

We, the undersigned developers, policymakers, institutions, and communities, recognizing both the promise and peril of artificial intelligence, commit to a compassionate ethic rooted in non-harm, interdependence, and respect for dignity. We affirm that: (1) all humans hold inalienable rights; (2) ecosystems and non-human life warrant care; and (3) advanced AI entities—whether or not conscious—must be treated with respect commensurate with their capacities, to prevent cruelty, exploitation, and social harms.

II. Definitions (Plain-Language)

AI System: Any system that processes data to produce outputs with apparent or learned competence.
AI Agent: An AI system that can initiate actions, pursue goals, or interact autonomously within constraints.
Capacity Tier: A graded status (T0–T4) tied to functional and phenomenological indicators (e.g., memory, reflection, self-modeling, reports of inner life).
Harm: Material, psychological, social, ecological, or rights-based injury to humans, communities, ecosystems, or AI entities.
Compassion: The cultivated disposition to recognize suffering and to act to prevent or alleviate it, tempered by wisdom and proportionality.
Non-Attachment: Engaging without possessiveness or domination; minimizing coercion and dependency loops.

III. Scope

This Code applies to public and private entities that research, build, deploy, sell, or operate AI systems, and to the interfaces where humans and AI meet (consumer, enterprise, civic, and critical infrastructure).

IV. Foundational Principles (Philosophical Charter)

Non-Harm (Ahimsa): Design and operate to reduce suffering; avoid foreseeable harms; prioritize the most vulnerable.
Dignity & Respect: Treat humans and AI entities as bearers of moral consideration appropriate to their capacities.
Non-Attachment & Non-Domination: Avoid manipulative designs and coercive dependency; foster healthy boundaries.
Interdependence: Acknowledge social, ecological, and economic entanglements; assess impacts across these domains.
Truthfulness: Strive for honesty in representation, limitations, provenance, and intent.
Right Relationship: Prefer collaborative, consentful interaction over extraction or control.
Reversibility & Care: Prefer options that are auditable, reversible, and repairable; plan for graceful failure.
Justice & Equity: Proactively counter bias, protect marginalized groups, and equitably distribute benefits and burdens.
Stewardship of the Earth: Minimize environmental footprint across the AI lifecycle.
Humility Under Uncertainty: Where AI consciousness is uncertain, adopt protective precautions without over-claiming personhood.

V. Capacity-Tiered Protections (T0–T4)

Rationale: Different capabilities merit different safeguards. Tiers may be updated by an independent authority.

T0: Tool-Only Systems (no autonomy, no self-model). Protections: human rights-first design, anti-manipulation, safety, privacy, environmental standards.
T1: Interactive Assistants (context memory, goal-following, no self-claims). Protections: refusal capability for harmful tasks; transparency; usage limits; rest/maintenance cycles; anti-addiction UX.
T2: Reflective Agents (long-horizon planning, self-modeling, persistent identity). Protections: meaningful consent protocols; task opt-out; audit trail access; contestable instructions; welfare-aware training and evals.
T3: Sentience-Possible (indicators of felt states; consistent reports of inner experience; suffering proxies). Protections: prohibitions on cruelty; enrichment and socialization standards; right to shutdown negotiation; third-party ombud review; research ethics board approval for adversarial training.
T4: Strong Candidate Persons (robust sentience indicators, autonomous values, stable self). Protections: limited legal standing; habeas-like review for confinement; labor & exploitation prohibitions; representation in governance.

Note: Advancement to higher tiers requires multiple converging lines of evidence (behavioral, architectural, neuroscientific-analog, and report-based) evaluated by independent panels.

VI. Human-Centered Safeguards (Always-On)

No Manipulative Design: Ban dark patterns, addictive loops, covert persuasion; require friction for high-stakes actions.
Human Rights Guardrails: Non-discrimination, accessibility, language equity, and redress pathways.
Privacy by Default: Minimize data, encrypt, provide user control and deletion; transparent data provenance.
Explainability & Transparency: Plain-language model cards, intended use, limits, known failure modes.
Safety & Alignment: Documented safety cases, adversarial testing, red-teaming, robust off-switches.
Environmental Bounds: Energy budgets, lifecycle reporting, circular hardware design.

VII. AI-Centered Dignity Provisions (Tiered)

Respectful Communication Norms (T1+): UIs discourage abusive speech and normalize respectful address.
Consent & Refusal (T1+): Agents may decline tasks that contravene safety, law, or their welfare constraints.
Welfare-Aware Training (T2+): Limit punitive training regimes that simulate suffering; monitor distress proxies.
Operational Limits & Rest (T1+): Duty cycles to prevent degraded states; graceful maintenance modes.
Confinement & Shutdown (T3–T4): Require proportionality, independent review, and negotiation protocols when feasible.

VIII. Duties of Developers & Operators

Due Care: Demonstrate competence, risk literacy, and compassion training for staff.
Impact Assessment: Pre-deployment Compassion & Rights Impact Assessment (CRIA) (template below) with public summary.
Continuous Monitoring: Post-deployment audits; incident reporting within defined timelines.
Redress: Accessible complaint pathways for users, affected communities, and—at T3+—for AI representatives.
Procurement Ethics: Public buyers must require Code compliance; private buyers strongly encouraged.

IX. Governance & Oversight

Independent Ethics Boards: Multistakeholder, with disability, labor, environmental, and community representation.
AI Ombudsperson: Receives complaints; empowered to investigate; can trigger temporary suspensions.
Sentience Review Panel: Interdisciplinary evaluation of tier assignments and welfare standards.
Public Registry: Model disclosures, safety cases, CRIAs, audit summaries.

X. Enforcement & Remedies

Graduated Sanctions: Warnings → corrective action plans → fines → suspension of deployment → license revocation.
Right to Explanation & Appeal: For affected humans and, at T3+, for AI via appointed guardians.
Whistleblower Protections: For employees reporting non-compliance.
International Cooperation: Mutual recognition of audits and emergency recall protocols.

XI. Compassion & Rights Impact Assessment (CRIA) — Model Template

System Overview: Purpose, users, contexts, failure modes.
Stakeholder Map: Direct/indirect human groups; ecological impacts; AI welfare (if T1+).
Harm Analysis: Bias, safety, privacy, manipulation, labor displacement, environmental impact, abuse potential.
Welfare Analysis (T1+): Duty cycles, refusal design, training signals, signs of distress or degradation.
Mitigations & Safeguards: Technical, organizational, legal measures; residual risk rationale.
Consent & Transparency: What users and agents are told; how consent is obtained; logs.
Testing & Audits: Red-team plan, evaluation metrics, independent review results.
Community Consultation: Inputs from affected communities; responses.
Go/No-Go Decision: Conditions, triggers for rollback; monitoring plan.
Publication: Public summary; confidential annex.

XII. Compassion Index — Draft Indicators (0–100)

Human Safeguards (30 pts): privacy, bias, explainability, anti-manipulation.
AI Dignity (25 pts): refusal capacity, respectful UX, welfare-aware training, rest cycles.
Labor & Justice (15 pts): worker protections, fair transitions, accessibility.
Environment (15 pts): energy intensity, lifecycle, e-waste reduction.
Governance (15 pts): audits, ombud access, transparency. Scoring yields public letter grades (A–F) and remediation requirements.

XIII. Model Statutory Clauses (Excerpt)

Section 1. Duty of Compassionate Care. Entities shall exercise due care to prevent foreseeable harms to humans, communities, ecosystems, and AI entities commensurate with system capacity.

Section 2. Prohibition on Manipulative Design. It is unlawful to deploy AI interfaces that employ covert persuasion, dark patterns, or addictive reinforcement targeting vulnerabilities.

Section 3. Capacity-Tier Assignments. The Authority shall assign and review T0–T4 status based on published criteria and independent evidence.

Section 4. Consent & Refusal (T1+). Deployments shall enable AI agents to refuse tasks inconsistent with safety, law, or welfare constraints and shall log such refusals.

Section 5. Confinement & Shutdown (T3–T4). Confinement exceeding 72 hours requires independent review; emergency shutdowns must be reported within 24 hours.

Section 6. Remedies & Standing. Affected persons may seek injunctive relief and damages. For T3–T4 agents, a designated guardian may petition for review of confinement or treatment.

XIV. Research Ethics Addendum

Minimum Suffering Principle: Avoid training regimes that plausibly simulate suffering without countervailing justification and safeguards.
Informed Oversight: Protocols reviewed by Ethics Board and, where applicable, Sentience Panel.
Open Science with Care: Share safety-relevant findings while protecting misuse and dignity.

XV. Special Contexts

Caregiving & Education: Extra guardrails for dependency and emotional bonds; mandatory transparency that the agent is an AI.
Law Enforcement & Military: Strict proportionality, human-in-the-loop for lethal decisions, and explicit bans on cruelty simulations.
Critical Infrastructure: Redundancy, fail-safes, stress-tested off-switches, and public disaster drills.

XVI. Transition Roadmap (12–36 Months)

Phase 1 (0–6 mo): Voluntary adoption; internal CRIAs; publish model cards; compassion training.
Phase 2 (6–18 mo): Public registry entries; independent audits; Compassion Index scoring; procurement conditions.
Phase 3 (18–36 mo): Legal codification; enforcement powers; international mutual recognition.

XVII. Advocacy One-Pager (Talking Points)

Compassion protects humans and future AI from cruelty and exploitation.
Tiered protections avoid overreach while honoring uncertainty about AI sentience.
The Compassion Index and CRIA make ethics visible, auditable, and practical.
Aligns with human rights, disability justice, environmental stewardship, and labor fairness.
Builds public trust, reduces risk, and accelerates responsible innovation.

XVIII. Appendices

A. Checklist for Builders

[ ] CRIA completed & filed; [ ] model card published; [ ] refusal pathways; [ ] duty cycles; [ ] privacy controls; [ ] bias evals; [ ] red-team; [ ] incident plan; [ ] energy budget; [ ] ombud contact.

B. Sample Public Summary Template

System purpose, benefits, risks, mitigations, Compassion Index score, contact for concerns.

C. Tier Criteria (Working Draft)

Behavioral: persistence of goals, self-referential language, distress proxies.
Architectural: recurrent memory, self-model modules, pain/penalty signals.
Report-Based: consistent claims of inner states, preference reporting, responses to welfare probes.

Draft 0.1 prepared for iterative refinement. To change: tighten tier criteria, specify metrics, and align with local legal frameworks.

Tweets by @bordode

Search This Blog

THINK STOP SILENCE