跳至正文
全部项目

商业摘要已提供本语言版本。产品详细流程和技术证据以英语维护。

安全与保障

Aegis AI

将 AI 清单、义务和证据纳入同一审查流程。

清单、分类、整改和报告位于不同位置。证据需要负责人和可审查的状态。

AI 产品负责人、合规负责人和承担责任的工程师

开发中的原型

A system evidence register
System
One representative AI use case
Ownership
Product, engineering and review owners
Gaps
Evidence and remediation to assign
Assessment
Qualified specialist review required
Decision logic and control flow

Explore the reasoning

Identify one AI system, its intended use and accountable owners.

Engage now
An evidence-workflow review covering ownership, gaps, remediation and specialist review needs.
Inspect
Development prototype
FastAPI modules and SQL migrations covering the principal workflow and data boundaries.
Still to validate
The test record relies substantially on local and mocked integrations.
Validate corpus currency, legal interpretation, integration behaviour, and end-to-end review with qualified compliance and system owners.

源码审查日期

The decision behind the system

Who owns the evidence behind this AI system?

A regulatory-workflow prototype that connects system inventory, retrieval, classification, remediation and reporting with accountable human review.

Aegis organises system-level governance and regulatory evidence. Achilles inspects code; EU AI Navigator explores early adoption and readiness.

Organise the review

Team assignments, role-based workflow surfaces, kanban and activity records give evidence work an owner and a visible state.

Connect the evidence

Retrieval, compliance snapshots, trends and AI Act–GDPR graph surfaces support structured review.

Prepare audit material

CSV and PDF exports turn reviewed workflow records into material for qualified assessment.

Read the complete system walkthrough
  1. Inventory

    Identify one AI system, its intended use and accountable owners.

  2. Examine

    Keep deterministic workflow records distinct from retrieval-assisted interpretation.

  3. Assign

    Link evidence gaps and remediation to named owners and source records.

  4. Review

    Qualified legal, compliance and system owners review the material. An export is not certification or a conformity assessment.

See all evidence and limits

Inspect the source, test conditions and operating limits in the technical record.

查看技术记录与证据

业务 ↔ 技术桥梁 · Governance capability experiment

Turn regulatory obligations into owned product and evidence work.

Explores evidence-linked regulatory workflow without presenting generated guidance as legal authority.

这是对项目组合与源代码的解读,并非客户、部署、认证或成果声明。

  1. 01

    业务压力

    System inventory, risk classification, evidence, remediation, and reporting are often fragmented across legal, compliance, product, and engineering teams.

  2. 02

    产品决策

    Separate deterministic inventory and workflow state from retrieval-assisted interpretation, and preserve named human review points.

  3. 03

    技术记录

    Inventory, retrieval, classification, evidence, remediation, audit, and report modules with migrations and selected mocked test seams.

  4. 04

    证据回流

    Evidence gaps and classification uncertainty return to product scope, control design, accountable owners, and release planning.

仓库/运行时边界

本页解读的是独立项目仓库中的源代码。源代码存在,并不表示该项目作为服务运行在 hyperion-consulting.io 内。

访客及客户数据的 AI 处理限于通过受治理的欧盟端点使用 Mistral。所有者授权的 Helio、搜索及 Product Council 流程已启用。独立的公开来源编辑流程使用 Anthropic,并以受治理的 Mistral 作为备用;该流程不接收网站咨询、客户文档或访客对话。远程 AI 不可用时,确定性工具仍可使用。

源码证据

源码实际能够证明什么

A regulatory-workflow backend prototype with inventory, retrieval, classification, evidence, remediation, audit, and report modules. It is a decision-support system under development, not legal advice or a conformity assessment.

详细技术快照以英文维护。

成熟度
Development prototype
证据审查
源码快照截至

产品与架构决策

  • Separate deterministic system inventory and workflow state from retrieval-assisted regulatory interpretation.
  • Link evidence chunks and report snapshots to the decisions they support.
  • Preserve human review points for classification, remediation, and regulatory reporting.

可核验的证据

  • FastAPI modules and SQL migrations covering the principal workflow and data boundaries.
  • Pytest coverage for selected API, data, classification, report, and remediation paths.
  • Regulatory-corpus ingestion and retrieval plans with mocked integration seams in tests.

声明边界

  • The test record relies substantially on local and mocked integrations.
  • No claim of complete regulatory coverage, legal accuracy, live monitoring, conformity, or certification is made.
  • An accountable legal, compliance, and system owner must review any generated assessment or document.

代码仓库证据不等同于客户、部署、认证或成果证据。

架构

从监管文本到体系化、以证据为支撑的报告

三个源代码层:监管信息流监测、检索辅助的义务提取,以及一条可检测篡改的审计链。监管机构是否认可任何一条记录,必须另行评估。

仅为源代码图:一层可能代表已实现代码、适配器边界或文档设计,并不证明已配置生产服务。

  1. 01

    第 1 层 — 监管监测

    源代码呈现三个 EUR-Lex ATOM 信息流(EU AI Act、GDPR、AI General)、按 CELEX 对新出台法律文书的解析、严重程度提取(critical/high/low),以及针对已登记 AI 系统清单的 Mistral 影响分析步骤。不声称存在实际运行的监测。

  2. 02

    第 2 层 — 合规引擎

    源代码呈现在条款、附件与 EDPB 指南语料库之上运行的 Mistral + RAG:义务提取、跨法规重叠映射(AI Act ↔ GDPR ↔ NIS2)、风险分类与差距分析,附带置信度分数和条款级引用。输出结果需要专家复核。

  3. 03

    第 3 层 — 可检测篡改的审计

    源代码在一条审计链中对分类、差距、整改、时间戳和复核人记录进行建模。导出的记录是否满足法律或监管机构的要求,必须另行评估。

EU AI Act 覆盖范围

原型中呈现的十项高影响力 AI Act 条款

合规表演只是把一份 PDF 里的条款列个清单。合规实务则是把每一条款映射到具体的 AI 系统,并配有证据、责任人和最近审查日期。下方列出该原型针对每个已登记系统所呈现的条款。提取出的义务需要专家复核。

仅为仓库表面:可用性、完整性、性能与运营适用性需要另行验证。

ART. 4

AI 素养

01

源代码包含按角色划分的素养跟踪、测验与记录生成工作流。它并不认证个人的能力,也不认证组织对第 4 条的合规。

ART. 5

被禁止的做法

02

一个规则与检索工作流将已登记的系统对照被禁止做法的类别与引用进行映射。法律上的分类仍然是由人作出的决定。

ART. 6

高风险分类

03

源代码呈现附件 III 风险等级评估,附带 Mistral 生成的推理过程、置信度分数以及按附件划分的义务树。其设计旨在于登记事件(更换模型、范围变更、训练数据更新)发生时重新运行。

ART. 9

风险管理体系

04

对风险进行持续的识别、估计、评估与处置。该原型的设计旨在将每个已识别的风险关联到其缓解控制措施,并从 CI 流水线中获取证据(测试通过/失败、模型卡、评测套件)。不声称存在实际运行的集成。

ART. 10

数据治理

05

具备代表性覆盖、偏差检测与数据血缘的训练、验证和测试数据集。该原型的设计旨在将数据集登记册与附件 IV 的技术文档要求进行交叉比对。

ART. 11

技术文档

06

附件 IV 技术档案:系统用途、设计、训练方法、性能指标。该原型的设计旨在生成文档骨架,从登记册中填入证据,并按附件 IV.1–9 标记缺失的章节。草稿需要由负有责任的人复核。

ART. 12

记录保存与日志

07

该原型呈现针对已登记系统的事件模式与审计记录检查。它并不证明外部的运行流水线会输出完整或在法律上充分的日志。

ART. 13

透明度

08

使用说明,包括预期用途、准确性、稳健性、可预见的误用以及人工监督措施。该原型的设计旨在对照第 13 条检查清单起草面向部署方的透明度告知。

ART. 14

人工监督

09

在系统设计中内置有效的监督措施。该原型的设计旨在将一项监督设计对照取自第 14 条的四项能力(理解、监控、推翻、停用)进行评估,并产出供复核的差距分析。

ART. 15

准确性、稳健性、网络安全

10

性能指标,对错误、故障、不一致情况的稳健性;以及抵御对抗性攻击的网络安全。该原型的设计旨在将评测套件结果、红队报告和 SBOM 汇集到附件 IV 的证据区块。不声称存在实际运行的集成。

技术栈

技术构建

所列技术仅标识仓库依赖项或适配器表面,并不证明存在活动集成或部署。

后端
Python, FastAPI, PostgreSQL, Redis, SQLAlchemy and Alembic
AI / NLP
Mistral Large (reasoning + classification), Mistral Embed (1024-dim vectors), Qdrant cosine similarity search, per-tenant vector collections
前端
Next.js, React, TypeScript, Tailwind CSS, React Flow and Recharts
数据管线
3 EUR-Lex ATOM feeds (AI Act, GDPR, AI General), PyMuPDF, python-docx, 1500-char semantic chunking, Pydantic structured extraction
基础设施
Docker Compose, S3/MinIO and custom JWT authentication; deployment requires separate validation
测试
Pytest for selected backend paths; frontend type checking and build records. This source review did not execute those checks.
Technical documents & controlled source access

源代码

查阅代码

出于评估与合作目的,Aegis AI 原型源代码可应请求提供。

需要签署 NDA

若要查阅源代码,请先签署保密协议(NDA)。这是一项标准的法律手续。

与 Hyperion 合作

审查一个 AI 系统的证据路径

审查证据工作流程,涵盖责任、缺口、整改及专业审查需求。

请简要说明您的情况、限制条件和待作决策。开始工作前,我们会商定范围、责任和报价。