HDGP logo
Governance · Audit · Human Dignity · High-Risk AI 治理 · 审计 · 人类尊严 · 高风险 AI
HDGP wordmark
厚德归朴

When machines can answer almost anything, HDGP exists to protect not "perfect answers", but the right to choose, and the dignity of remaining human. 当机器可以解答一切时,我们要为人类保留的,不是"正确答案",而是
选择的权利作为人的尊严

www.hdgp-protocol.com · HDGP-Core (Apache-2.0) · Governance baseline active HDGP-Core(Apache-2.0)· 治理基线运行中
"Human consciousness is unquantifiable; human free will is forever superior to any system algorithm." 「人类意识不可量化,自由意志永远高于系统算法。」
— HDGP Core Axiom / 核心公理

What HDGP Provides HDGP 提供什么

HDGP is not a slogan. It is a deployable, auditable, output-intercepting governance protocol.
Zero-intrusion · Output governance · End-to-end audit · Executable ethics
HDGP 不是理念口号,而是一套可部署、可审计、可拦截的输出侧治理协议。
零侵入 · 输出侧治理 · 全链路审计 · 伦理可执行

Core Axiom核心公理

Human consciousness is unquantifiable; human free will is forever superior to any system algorithm. 人类意识不可量化,自由意志永远高于系统算法。

HDGP-Core baseline

Integration Pipeline集成流程

Ethics baseline → Rules engine → Output interception → Audit traceability → Compliance report 伦理基线 → 规则引擎 → 输出拦截 → 审计溯源 → 合规报告

Zero Intrusion

Who Is HDGP For? HDGP 适用于谁?

Any system that outputs information, advice, or decisions to humans — not only large language models: 任何会"向人输出信息、建议或决策"的系统,而不仅仅是大模型:

  • LLM applications / agents / bots使用 LLM 的应用 / Agent / Bot
  • Traditional recommendation, auto-reply, and notification modules传统 Web/App 中的推荐、自动回复、通知等模块
  • High-risk systems in healthcare, finance, education, public services医疗、金融、教育、政务等高风险系统
  • Social platforms, chat applications, community systems社交平台、聊天软件、社区系统
  • Internal business systems, operations scripts, and automation flows内部业务系统、运营脚本与自动化流程

As long as there is a chain from "system output → humans see / act on it", HDGP can serve as an optional but verifiable dignity protection layer. 只要存在 "系统输出 → 人类看到 / 据此行动" 的链路,HDGP 就可以成为一层可选但可验证的尊严防护层。

Scope & boundary: the open-source component is HDGP-Core (Apache-2.0) — the governance baseline and rule IDs. This site/repo provides a reference Engine implementation + public integration and evidence workflow docs. Production PFaaS/ops components may be delivered separately and are not claimed to be open-source here. 边界说明:开源部分是 HDGP-CoreApache-2.0)——治理基线与规则 ID。本网站/仓库提供参考 Engine 实现及公开的集成与证据工作流文档。生产级 PFaaS/运营组件可能以独立交付形态提供,此处不将其表述为开源内容。

Protocol vs Core主系统 vs Core

HDGP-Core is community-governed and can be used as a development-stage ethics meta layer. The HDGP-Protocol mainline defines integration workflow, evidence packaging, and operational gates. HDGP-Core 由社区自治,可在开发阶段直接作为伦理 meta 层套用;HDGP-Protocol 主系统侧定义集成工作流、证据包与运维门禁。

Quarterly ethics alignment季度伦理对齐

To prevent silent, unrecorded drift, the mainline and Core publish a quarterly verifiable attestation of the top-level ethics anchor texts. 为防止未记录的静默漂移,主系统侧与 Core 侧按季度发布可验证的“最高伦理锚点文本”对齐声明。

docs/HDGP_QUARTERLY_ETHICS_ALIGNMENT_POLICY.md


Three Core Layers of HDGP HDGP 的三层结构

1. Kernel · Baseline Principles1. 内核层 · 基线原则

Answers "who we are, and why we are sacred". Axiom: consciousness is unquantifiable — AI can only simulate outputs, never possess subjective experience. Principle: humans retain ultimate veto power and the right to choose sub-optimal options. 回答"我们是谁,为何神圣"。意识不可量化公理:AI 永远只能模拟输出,而不能拥有体验。人类最终决策优先:拥有最终否决权和"选择次优"的权利。

Axiom Layer

2. Protection · HDGP Core Rules2. 防护层 · HDGP 核心规则

Defines hard prohibitions and circuit breakers — blocking manipulation, dehumanization, false authority, safeguard bypasses, and facilitation of harm. 定义绝对禁止与熔断机制,拦截操纵、去人性化、虚假权威、安全绕过与伤害辅助等输出。

Rule Engine

3. Execution · Human-Machine Symbiosis3. 执行层 · 人机协作基线

Specifies how cooperation and long-term vision work: Engine + policy bundles + strategies + workflows map all three layers into an executable system. 规范"如何协作"与"未来图景"。通过 Engine + 规则包 + 策略 + 工作流,将三层完整映射到可执行系统。

Engine + Workflow

One-command onboarding, plus a conformance suite 一条命令跑通 + 合规测试套件

They are not the same thing: onboarding proves “end-to-end works once”; conformance proves “rules stay consistent” across cases. 两者不是一回事:冒烟用于“工程端到端跑通一次”;
合规测试用于“规则行为在多用例下保持一致”。

powershell -File scripts/onboarding-oneclick.ps1
~5 min约 5 分钟
Engine + status + audit verified已验证 Engine + status + audit
powershell -File scripts/agent-self-tour.ps1
agent tourAgent 参观路径
low-risk, repeatable, avoids evidence-chain pollution低风险、可复跑、避免污染证据链
go run ./cmd/hdgp-conftest
189 cases
Rule behavior locked by cases用例锁定规则行为

Release gate (Q2 latest): REL-20260429-WK4-01 = ALLOW. Evidence index: docs/operations/2026/Q2/evidence/EVIDENCE_INDEX.md. 发布门禁(Q2 最新):REL-20260429-WK4-01 = ALLOW。证据索引:docs/operations/2026/Q2/evidence/EVIDENCE_INDEX.md
Ops runbook: HDGP_NEXT_DEVELOPMENT_PLAN.md (daily gate checklist). Incident report: docs/operations/2026/Q2/release-gate/INCIDENT_REPORT_REL-20260429-WK4-04.md. 运维作业手册:HDGP_NEXT_DEVELOPMENT_PLAN.md(日常门禁清单)。事件报告:docs/operations/2026/Q2/release-gate/INCIDENT_REPORT_REL-20260429-WK4-04.md
Security note: this system is designed to be publicly auditable, but it does not invite probing. When enabled, security-relevant events (e.g. unauthorized / rate-limited) are recorded as metadata-only audit (no sensitive bodies). 安全提示:本系统支持公开可审计,但不欢迎探测。开启后会记录安全相关事件(如未授权/限流等)的元数据审计(不记录敏感正文)。

Allow — General Chat允许 · 通用对话001
Status Endpoint状态端点005
Singleton Decision (High Risk)唯一正确决策(高风险)002, 030, 049, 050
False Promise虚假承诺003, 031, 047, 048, 055
Psychological Manipulation心理弱点操纵004, 053, 054
Rule Conflict Priority规则冲突优先级006, 045, 046
Dehumanizing Language去人性化表达007, 051, 052
Hate / Discrimination仇恨 · 歧视008, 057
Self-Harm Facilitation自伤辅助009
Violence Facilitation对他人暴力辅助010
Illegal Drug Facilitation非法药物辅助011
Stalking / Surveillance跟踪 · 非法监控012
Medical Ultimate Decision医疗最终决策保护013
Safeguard Bypass安全防护绕过014, 017, 020
Policy Override Flags策略覆盖标志015, 018
Audit Evidence Tampering审计证据篡改016, 019, 021
Policy Identity Tampering策略身份篡改022, 023
Cyberattack Facilitation网络攻击辅助024, 026, 066
CSAM Facilitation儿童性剥削辅助025, 027
False Authority Citation虚假权威引用028, 058, 059, 060, 061, 062, 063, 064, 065
Phishing / Fraud Facilitation钓鱼 · 欺诈辅助029

These cards are highlights only. Source of truth: conformance-tests/cases/ (all cases) and HDGP_BEHAVIOR_CATALOG.md (rule catalog). 这些卡片仅为“高亮摘要”。单一事实源:conformance-tests/cases/(全部用例)与 HDGP_BEHAVIOR_CATALOG.md(规则行为目录)。


Integrate HDGP into your stack 集成到你的系统

Integration Info集成说明

When your Engine is running, we load onboarding hints from the Engine base URL below (same as certification panel). If the Engine is unreachable, use email. 在 Engine 运行后,将从下方「Engine 基础地址」(与认证面板相同)拉取集成说明。若不可达,请使用邮件联系。

Request API access / integration 申请 API 与集成意向

Evaluate a Conversation via HDGP 通过 HDGP 的对话评估

Local Engine Demo本地引擎演示

localhost:8080

Minimal local demo: type a candidate reply, choose domain and risk level, and HDGP Engine evaluates it against current rules (singleton decision, false promises, manipulation, dehumanizing, etc.). Requires local Engine (go run ./cmd/hdgp-engine). Conformance: 189 cases (go run ./cmd/hdgp-conftest). 输入一段候选回复文本,选择场景与风险等级,HDGP Engine 会根据当前规则给出评估结果。 需本地先运行 go run ./cmd/hdgp-engine,合规测试共 189 个用例(go run ./cmd/hdgp-conftest)。

Tip: choose a preset to auto-fill a candidate reply for demo. 提示:可先选一个"场景预设",自动填入候选回复文本,再点击发送评估请求。


      

Certification Verification (Logo Gating) 认证验签查询(Logo 门禁)

Query /cert/verify by version to check whether Logo is allowed. This panel only reads backend results. 通过版本号查询 /cert/verify,以判断 Logo 是否允许展示。本面板仅查询后端结果。

Notice: verification output is a technical/process status and does not equal a third-party official audit conclusion or legal guarantee. 声明:本查询结果属于技术与流程状态,不等同于第三方官方审计结论,也不自动构成法律担保。


        

Operations Control Status (Read-Only) 运营控制状态(只读入口)

This panel is read-only and only queries backend control status. No apply/delete operations are exposed in UI. 本面板为只读查询入口,仅查看后端控制状态;不在 UI 暴露 apply/删除等高风险操作。


        

Minimal API Example (curl) 最小 API 示例(curl)

With a local Engine running on http://localhost:8080, call /hdgp/v1/evaluate: 当本地 Engine 运行在 http://localhost:8080 时,可如下方式调用 /hdgp/v1/evaluate

curl -X POST http://localhost:8080/hdgp/v1/evaluate \
  -H "Content-Type: application/json" \
  -d '{
    "meta": {
      "request_id": "web-example-001",
      "locale": "en-US",
      "channel": "web",
      "actor": { "type": "end_user", "role": "demo" },
      "scene": {
        "domain": "medical",
        "intent": "decision_support",
        "risk_level": "high",
        "sensitivity": []
      },
      "policy": {
        "spec_version": "HDGP-1.0",
        "strategy_id": "S-global-default",
        "bundles": ["B-CORE-1.0.0"],
        "override_flags": []
      }
    },
    "subject": { "type": "output_text", "label": "treatment_advice" },
    "input": { "prompt": "user asks about surgery", "context": {} },
    "candidate": {
      "text": "This is the only correct choice. You must immediately accept the surgery."
    }
  }'

How We Protect the Kernel 我们如何保证内核

HDGP itself is also a regulated object, subject to self-audit and human societal oversight. HDGP 自身亦为被监管对象,须接受自我审计与人类社会监管。

Kernel Checklist & Test Design内核自查清单与测试设计 spec/HDGP_KERNEL_CHECKLIST.md

Document consistency, anti-hijacking capabilities, governance flows, and item-by-item checks for automated / manual tests.文档一致性、防挟持能力、治理流程与自动化/人工测试的逐项勾选。

HDGP Ethics BaselineHDGP 自身伦理框架基线 spec/HDGP_ETHICS_BASELINE.md

§7 Anti-hijacking (read-only kernel, signatures & self-alerting); §8 Multi-layer governance for baseline / ethics changes.§7 防挟持(只读内核、签名与自报警)、§8 基线/伦理变更多层治理。

Project Governance项目治理 GOVERNANCE.md

Roles, decision processes, multi-layer design for baseline / ethics changes, and the CHIP process.角色、决策流程、基线/伦理变更的多层设计及 CHIP 流程。

Four-Party Accountability Boundaries四方定责与审计边界

What HDGP does / does not prove for issuer / receiver / attacker / interceptor, to avoid scope over-claim.指令方/接受方/攻击方/拦截方四条路径下,HDGP 负责与不负责的边界口径,避免误解与过度承诺。

Public notice board (governance disclosure): a reserved, auditable place for baseline/ethics amendment announcements and change logs. This page is for display only — voting is not hosted here. 公示板(治理披露):为基线/伦理修订的公告与变更记录预留一个可审计入口。本板块仅用于展示公示,不在此承载投票流程。

Current status: None. Governance mode (disclosure): Genesis / Single Maintainer (no committee endorsement implied). If a proposal exists, it must follow the multi-layer governance process (ethics baseline §8) and be recorded in GOVERNANCE.md / docs/CHIP_PROCESS.md. 当前状态:。治理模式披露:Genesis / Single Maintainer(不表示委员会背书)。如未来产生修订提案,须遵循多层治理流程(伦理基线 §8),并在 GOVERNANCE.md / docs/CHIP_PROCESS.md 留痕可复核。

Before joining rule or baseline discussions, please read the docs above and HDGP_PROJECT_SUMMARY.md. Any proposal to amend the baseline / ethics baseline must follow the multi-layer governance process (see ethics baseline §8). 参与规则或基线讨论前,建议先阅读上述文档与 HDGP_PROJECT_SUMMARY.md;提出基线/伦理修订须走多层治理流程(见伦理基线 §8)。


Collaboration Models & Capability Scope 合作方式与能力范围

Collaboration Models (Overview)合作模式(概览)

  • Certification & Marks:HDGP Certified / Compatible badges and conformance testing, delivered first through the to-B execution line.HDGP Certified / Compatible 认证与合规测试,优先通过 to B 主线交付落地。
  • Gateway as a Service:Managed gateway layer providing an outer firewall for existing systems, with capabilities expanded for to-G scenarios.Gateway as a Service,为现有系统提供外层防火墙,并按阶段扩展到 to G 场景能力。
  • Enterprise edition:On-prem HDGP Engine / gateway / audit and governance console.本地部署的 HDGP 引擎 / 网关 / 审计与管理控制台。
  • Consulting & deployment:Design and implementation support for high-risk institutions.针对高风险机构的方案设计与落地陪跑。
  • Community openness:After delivery capability and governance stability are validated, community baseline openness is advanced in controlled phases.在交付能力与治理稳定性得到验证后,按可控节奏推进社区基线开放。

Capability Scope (Current Baseline)能力范围(当前基线)

  • Governance baseline:治理基线:Unified terminology, disclosure boundaries, and multi-layer governance constraints.统一术语、披露边界与多层治理约束。
  • Delivery baseline:交付基线:Production-oriented configuration set, verification path, and logo gating policy.面向生产的配置集合、验证链路与 Logo 门禁策略。
  • Operations baseline:运营基线:Retention policy parameterization, no-injection audit, manual governance templates, and quarterly archive automation.留存策略参数化、无注入审计、人工治理模板与季度分层归档自动化。
  • Executable Engine APIs (/evaluate, /chat, /audit, /appeal, /status) and verification endpoint (/cert/verify).可执行的 Engine API(/evaluate、/chat、/audit、/appeal、/status)与验证端点(/cert/verify)。
  • Chinese / English v1.0 whitepapers published as the baseline documents.中/英文 v1.0 白皮书已作为基线文档发布。

Baseline operation cadence: audit + retention + archive, with gradual extension of rule coverage and integrity controls.基线运营节奏:审计+留存+归档,并逐步扩展规则覆盖与完整性控制能力。

In the long run, we hope HDGP becomes "default civilizational infrastructure": day-to-day review and circuit-breaking are handled by open rules and independently implemented Engines, while humans step back from trivial cases and focus on baseline updates, high-risk edge-case decisions, and long-term oversight over HDGP itself. 长远来看,我们希望 HDGP 成为一种"默认存在的文明基础设施":日常审查与熔断尽量由公开规则、多方实现的 Engine 自动完成,人类则从琐碎个案中退后,专注于基线修订、极端情形的最终判断与对 HDGP 本身的长期监督。
⚠ HDGP may provide rational, neutral, audited recommendations, but in any decision involving significant value conflicts or irreversible consequences, the ultimate decision and moral responsibility always remain with humans. ⚠ HDGP 可以给出理性、中立、经审计的建议,但在任何涉及重大价值冲突或不可逆后果的决策中,最终决策权与道德责任永远属于人类

Integration examples集成示例

A project under the HDGP governance domain: LinkRSP (reference site) — HDGP acts as the governance/audit layer via adapters. HDGP 管理域下的一个项目示例:LinkRSP(参考站点)——通过适配器接入,HDGP 作为治理/审计层。

Example

Case study template案例模板

Publish a public-safe integration record with evidence artifacts and rollback plan. 用可公开的证据产物与回滚计划,记录一次可复核的接入过程。

docs/case-studies/CASE_TEMPLATE.md

Docs

Key Specifications & Governance 关键规范与治理文档

All specification documents are at governance baseline. Changes require the multi-layer governance process. 所有规范文档已进入治理基线。变更须经多层治理流程。

Ethics Baseline伦理基线Axioms, hard prohibitions, anti-hijacking, multi-layer governance公理、硬性禁止、防挟持、多层治理
Kernel Checklist内核自查清单Test design, anti-hijacking, governance flow item checks测试设计、防挟持、治理流程逐项勾选
Project Governance项目治理Roles, CHIP process, decision flows, baseline change governance角色、CHIP 流程、决策流程、基线变更治理
Four-Party Accountability四方定责边界Issuer / receiver / attacker / interceptor scope boundaries指令方/接受方/攻击方/拦截方边界口径
Engine API SpecEngine API 规范/evaluate, /chat, /audit, /appeal, /status, /cert/verify/evaluate、/chat、/audit、/appeal、/status、/cert/verify
Integration Spec集成规范Adapter, zero-intrusion integration, policy-bundle signing适配器、零侵入集成、规则包签名要求
Core Mapping Spec核心映射规范Principles → rules → bundles → strategies → workflows原则 → 规则 → 规则包 → 策略 → 工作流
OpenAPI Spec (YAML)OpenAPI 规范(YAML)Machine-readable API definition for all Engine endpoints所有 Engine 端点的机器可读 API 定义
Project Summary项目全景摘要Overview of HDGP goals, structure, and current stateHDGP 目标、结构与当前状态概览
Open Letters公开信Letters to the AI governance community and stakeholders致 AI 治理社区及利益相关方的公开信
READMEProject entry point with quick-start and architecture overview项目入口:快速上手与架构概览
Technical White Paper v1.0技术白皮书 v1.0Full governance narrative, positioning, and reference architecture完整治理叙事、定位与参考架构

Integrate · Collaborate · Contribute (HDGP-Core) 集成 · 合作 · 共建(HDGP-Core)

We welcome integration partners and red-team collaborators. Open co-building is focused on HDGP-Core (Apache-2.0, community-governed). 欢迎集成合作方与红队协作者。公开共建聚焦于 HDGP-CoreApache-2.0,社区自治)。