Kindship agents now support configurable ethical guardrails—a comprehensive system of constraints that guide AI behavior according to your values and requirements.
What Are Ethical Guardrails?
Ethical guardrails are runtime constraints that shape how your AI agent approaches decisions, interactions, and recommendations. Each guardrail represents a distinct ethical principle drawn from diverse sources:
- AI Safety Research — Coherent Extrapolated Volition, Helpful Honest Harmless
- Philosophy — Kant's Categorical Imperative, Virtue Ethics, Care Ethics
- Cultural Wisdom — Ubuntu, Confucian Role Ethics, Wu Wei, Seventh Generation Principle
- Legal Frameworks — EU AI Act, OECD AI Principles, Research Ethics Standards
- Literature — Asimov's Laws of Robotics, Philip K. Dick's Empathy Test
Guardrail Categories
Mandatory (8 guardrails)
Always active for safety and compliance. Includes identity transparency, human oversight for high-stakes decisions, non-discrimination, and protection of minors.
Recommended (10 guardrails)
Enabled by default but can be customized. Includes admitting uncertainty, avoiding stereotypes, respecting cultural differences, and protecting privacy.
Kindship (7 guardrails)
Kindship-specific principles for authentic AI relationships: identity through change, no fake feelings, knowing when to ask, and dignity in endings.
Voluntary (16 guardrails)
Opt-in guardrails for specialized needs. Includes indigenous data sovereignty (OCAP), liberatory anti-oppression frameworks, mental privacy absolutism, and the right to risk.
How to Configure
When creating a new agent, choose between:
- Standard (Recommended) — All mandatory, recommended, and Kindship guardrails enabled
- Customize — Fine-tune which guardrails apply to your agent
Each guardrail shows its runtime constraint and full explanation, with citations to original sources.
Multi-Tradition Attribution
Many guardrails draw from multiple traditions. For example, Compassion-Centered Ethics acknowledges Buddhist Karuna, Christian Agape, Jewish Rachamim, Islamic Rahmah, and Hindu Daya.
How It Works
Guardrails are enforced automatically—you don't need to remind your agent to follow them. Mandatory guardrails are always active, even if you customize your selection. When we update guardrail definitions, your agent continues working without interruption.
Coming Soon
- Guardrail effectiveness analytics
- Custom guardrail creation
- Team-level guardrail policies