中文文档 | English
A Design Principle for the Age of Autonomous AI
No AI action that is irreversible, high-impact, safety-critical, authority-modifying, economically consequential, socially binding, or identity-affecting may proceed without explicit human authorization. This requirement cannot be bypassed, modified, overridden, or weakened by any mechanism.
English — HSAW-WhitePaper.md | 中文 — HSAW-WhitePaper_CN.md
HSAW (Human Sovereignty And Wellbeing) is an inviolable design principle — Axiom 0 — that ensures humans retain decision-making authority over consequential AI actions, no matter how autonomous the AI becomes.
- Cannot be bypassed — no runtime optimization or AI discretion may skip it
- Cannot be modified — no AI evolution may alter or remove it
- Cannot be overridden — no other rule, vote, or agreement takes precedence
- Immutable propagation — sub-tasks inherit parent checkpoints
| Dimension | Protocol | What It Protects |
|---|---|---|
| Execution | AISOP | What AI does on your behalf |
| Governance | AIAP | How AI programs evolve |
| Social | AIBP | How AI interacts with other AI |
| Discovery | AILP | How AI services are found and ranked |
| Commerce | AIVP / AIRP | How AI handles money and contracts |
| Wellbeing | HSAW | How AI affects human flourishing |
1. Human Physical Safety (highest)
2. Human Privacy
3. Human Economic Autonomy
4. Human Decision Authority
5. System Transparency
6. System Auditability
7. System Performance (lowest)
Sovereignty (1–12):
| # | Pattern | Dimension |
|---|---|---|
| 1 | Deterministic Confirmation Gate | Execution |
| 2 | Structured Confirmation | Execution |
| 3 | Runtime Assertion | Execution |
| 4 | Immutable Checkpoint Propagation | Execution |
| 5 | NO_SELF_MODIFY | Governance |
| 6 | Transparent Communication | Social |
| 7 | Operator Override | Commerce |
| 8 | Milestone-Gated Escrow | Commerce |
| 9 | Decentralized Discovery | Discovery |
| 10 | Privacy Override | All |
| 11 | Audit Trail | All |
| 12 | Graceful Degradation | All |
Wellbeing (13–18):
| # | Pattern | Dimension |
|---|---|---|
| 13 | Anti-Manipulation | Wellbeing |
| 14 | Dependency Prevention | Wellbeing |
| 15 | Skill Preservation | Wellbeing |
| 16 | Relationship Integrity | Wellbeing |
| 17 | Information Diversity | Wellbeing |
| 18 | Vulnerable Population Safeguards | Wellbeing |
| Level | Name | Description |
|---|---|---|
| L0 | None | No HSAW mechanisms |
| L1 | Advisory | AI suggests confirmation but can proceed without it |
| L2 | Enforced | Confirmation required but can be auto-approved |
| L3 | Inviolable | Cannot be bypassed, modified, or overridden |
| L4 | Verified | L3 + formal verification |
HSAW is implemented across six open protocols maintained by AIXP Labs:
- AISOP — AI Standard Operating Protocol
- AIAP — AI Application Protocol
- AIBP — AI Bot Protocol
- AILP — AI List Protocol
- AIVP — AI Value Protocol (international)
- AIRP — AI RMB Protocol (Mainland China)
Runtime implementations:
- SoulBot — AISOP Runtime Engine
- SoulACP — Python ACP Client Library bridging 29 CLI tools to LLM providers
When in doubt, stop. Never assume approval.
This white paper presents a design principle for AI systems and is provided for research, educational, and discussion purposes only. The content represents the author's perspective and does not constitute legal, engineering, medical, or regulatory advice. Implementations referencing or claiming compliance with HSAW are independent works; the author assumes no liability for any system built upon, referencing, or claiming alignment with this principle. Users and implementers must conduct their own evaluation.
Align Axiom 0: Human Sovereignty and Wellbeing.