跳到正文
原文
欧盟 AI Act 官方·· 20 天前AI 评分46

欧盟委员会发布内部 AI 使用框架,GPT@EC 代码计划 2026 年开源

The Commission internal use of Artificial Intelligence

文章要点
  • 1欧盟委员会发布内部 AI 使用文件,为员工提供覆盖生成式 AI、采购与网络安全的指南
  • 2内部 AI 登记册按与 AI Act 对齐的风险方法为系统分类,用于合规跟踪和报告
  • 3内部生成式工具 GPT@EC 面向全体员工开放并提供 API,代码预计 2026 年开源
  • 4AI4DRPM 系统为立法草案生成数字声明草稿,政策官员保留最终控制权
决策要点

欧盟委员会以自身为样板,展示大型机构如何把 AI Act 的风险分级要求落进内部 AI 管理与员工工具使用。

  • 影响:与欧盟机构有业务往来的企业将面对按 AI Act 风险分级的客户治理与采购要求;GPT@EC 代码计划 2026 年开源,机构级生成式 AI 的私有化部署可能多一个参照。
  • 参考:可考虑参照其做法盘点自家 AI 系统清单并按风险分级登记,作为应对 AI Act 的内部底稿;值得跟踪 GPT@EC 开源进展,评估私有化部署的适用场景。
  • 依据:内容为欧盟委员会官方发布,属官方口径;内部工具效果与 2026 年开源时间均为官方表述,暂无独立验证。
正文

DIGITAL SERVICES ACT

The Commission internal use

of Artificial Intelligence

October 2025

Leading by example: The European Commission is striving to become a modern, responsible and adaptive

institution by implementing robust AI initiatives under the AI@EC Communication. These initiatives ensure

that AI development and use within the Commission are lawful, safe, and trustworthy.

Driving responsible AI uptake

• Clear Guidelines: The Commission has developed comprehensive guidelines for AI use, development,

procurement, and cybersecurity for its Staff. This includes resources on generative AI, data management,

supercomputing, sustainability, and open-source development, alongside practical tools like FAQs on

intellectual property and an AI guide for managers.

• Risk-Based AI Register: AI systems in development or use are classified based on a risk-driven

methodology aligned with the AI Act, tracked in an internal register to ensure compliance and facilitate

reporting.

• Ongoing Oversight: The Interservice Steering Group on AI regularly monitors internal AI initiatives, such

as the development of new tools, training, corporate communication and guidelines, ensuring they align

with the Commission’s goal of fostering the use of safe and trustworthy AI.

Empowering an AI-Skilled Workforce:

• Training and Community: The Commission fosters AI literacy through training programmes, the AI@

EC portal for knowledge sharing, the AI@EC Network for community practice, and peer learning via the AI

Champions network. An internal survey helps identify gaps in AI tool adoption.

Some examples of how AI helps Commission Staff:

• Support with repetitive or administrative tasks .

• Support evidence-based policy making by helping staff navigate vast amounts of information.

• Support multilingualism .

GPT@EC: An in-house generative AI tool

available to all staff, with API access to enhance

other systems. Its code is expected for a first

open-source release in 2026.

Language tools such as e-Translation and others,

provide new capabilities thanks to new generative AI

features.

AI4DRPM (AI for Digital-Ready Policy-Making):

Digital statements are now attached to every

legislative proposal. AI4DRPM is a system analysing

draft legislation, detecting and classifying digital

relevance and producing a draft digital statement

expected for every legislative proposal. Policy officers

remain in full control, finalising the output.

AI

来源:欧盟 AI Act 官方 · digital-strategy.ec.europa.eu