内容来源：https://aibusiness.com/ai-policy/how-ai-reshaping-europe-s-digital-sovereignty-debate

内容总结：

欧盟报告首提“AI主权”：各国需在数字化转型中加速布局

由谷歌云赞助的一项研究显示，各国政府在布局生成式人工智能时，应优先聚焦能改善人类信息体验的领域。与此同时，欧盟委员会支持的一项电子政务基准报告警告称，欧盟成员国需加快数字化转型步伐。

这份上周发布的年度报告首次将“人工智能主权”纳入评估体系，并将其列为欧洲关键技术优先事项之一。报告指出，AI主权的提出标志着各国对技术控制权的思考正在转变——政府关注的焦点已从数据存储地延伸至模型与基础设施的搭建方式，以及如何确保对这些技术应用的监督。

法国IT咨询公司凯捷执行副总裁兼公共部门负责人、报告撰稿人马克·莱因哈特在访谈中强调，AI主权正快速成为政策讨论的核心议题。他指出，过去关于主权的讨论多集中在基础设施层面，如云服务商、数据托管地点等；而如今，随着欧盟通过新政策举措明确数字主权的实际操作定义，讨论范围已扩展至整个数字技术栈。

“AI系统并非孤立存在，它运行于基础设施和模型之上，并嵌入特定假设、价值观和推理方式。”莱因哈特表示，各国不仅关注数据存放位置，更开始追问使用了哪些模型、由谁开发、决策逻辑是否透明。对于社会福利裁定、敏感政府服务等场景，这些问题尤为关键。

在实践中，AI主权并非单一解决方案。莱因哈特强调，从不同云环境到主权AI平台及本地化部署，各类选项的“主权级别”应取决于工作负载的敏感性和风险程度。例如，公开的商业注册数据与医疗健康记录所需的安全级别显然不同。

针对追求主权是否会牺牲AI创新的担忧，莱因哈特认为两者并非“二选一”。正如组织越来越多地采用多云环境，未来也将进入“多AI”世界，不同类型模型服务于不同目的。他坦言：“我们必须解决问题，改善社会和民众生活。如果欧洲现有方案无法做到，组织就需要务实，利用市场上可得且负担得起的技术。”

以德国联邦数字化与现代化部合作的主权AI平台为例，莱因哈特指出，项目启示在于“应从要解决的问题出发，而非为追求主权而追求主权”。该项目为重大基础设施规划审批流程开发了解决方案，同时采用了欧洲AI替代方案等高主权组件，证明欧洲技术在许多场景下已能交付可靠成果。

莱因哈特最后建议，政府需改革传统技术采购与部署模式。过去基于长周期规划、详尽规格和数十年系统寿命的路径已不适用于快速迭代的AI技术。政府应转向更短的开发周期、更小规模的项目，并持续评估技术选项。“公共部门向来以规避风险为设计原则，但AI需要更多实验精神。有时，过于求稳反而无法带来实际效益。”

中文翻译：

由谷歌云赞助
选择您的首批生成式AI应用场景
要开始使用生成式AI，首先聚焦于能够改善人类信息体验的领域。

各国政府更有效追求AI主权的策略。

一份由欧盟委员会支持的电子政务标杆报告警告称，欧盟成员国需加快数字化转型步伐。
上周发布的这份年度报告将AI主权列为欧洲关键技术优先事项之一。
首次纳入标杆体系的AI主权，反映出各国政府思考技术控制方式的转变。政府不再仅关注数据存储地点，而是开始关注模型与基础设施的实际构建方式，以及如何维持对其使用的监督。

在本问答中，法国跨国IT咨询公司凯捷执行副总裁兼公共部门负责人、该报告的撰稿人之一马克·莱因哈特，探讨了AI主权日益增长的重要性，以及为何在AI时代政府可能需要重新思考技术采购与部署的传统方法。

为何您认为主权会被纳入今年的报告？
马克·莱因哈特：这无疑是报告的新阶段。主权出现在标杆体系中本身就意义重大。历史上，关于主权的讨论聚焦于基础设施——云服务商、数据托管以及关键系统的所在地。如今，随着欧盟通过新政策举措与要求来定义数字主权的实际内涵，讨论范围正在扩大。
在此背景下，各国政府自然开始评估自身立场。他们追问的是：整个数字堆栈中的主权意味着什么？他们需要对支撑公共服务的各项技术拥有何种程度的控制权？

AI在更广泛的数字主权讨论中处于何种位置？
莱因哈特：AI为这场辩论增添了全新维度。AI系统并非孤立存在——它们运行在基础设施与模型之上，并内嵌了假设、价值观与推理方式。
各国不仅关注数据存储地点，也愈发追问：他们使用何种模型？由谁开发？是否理解这些系统基于训练数据做出决策的方式？对某些应用场景而言，这些问题或许无关紧要；但对另一些场景，如福利决策或高度敏感的政府服务，它们变得至关重要。
与相对成熟的云主权相比，AI主权仍在演进，其定义也更具流动性。但它正迅速走向政策讨论的前沿。

AI主权在实践中如何体现？
莱因哈特：围绕主权的讨论通常带有政治动机，要求有时非常抽象。如今，我们正致力于澄清并揭开AI主权的神秘面纱。
起点是理解你试图保护什么以及为何保护。是担心数据访问？系统韧性？监管合规？还是国家安全？一旦明确了目标，识别合适方案就变得容易得多。
现实是，主权并非单一的解决方案。从不同的云环境到主权AI平台及本地控制部署，有广泛的选择范围。今天的挑战不再是缺乏选项，而是理解哪种选项适用于哪种场景。
适合公开商业注册数据的方式，可能与健康记录或其他敏感信息的要求截然不同。主权的层级应反映工作负载的敏感度及所涉风险。

追求主权是否可能以牺牲AI创新为代价？
莱因哈特：确实存在权衡，但正在形成的共识是：这并非非此即彼的选择。
正如组织日益在多云环境中运营，我们很可能走向一个多AI的世界——不同模型用于不同目的。某些应用可能使用领先的商业模型，而另一些可能依赖开源或本地托管系统，以实现更强控制。
目标本质上是寻找主权与解决实际问题之间的平衡。归根结底，我们必须解决问题。我们必须找到改善社会与民众生活的方案。如果欧洲现有的解决方案无法做到这一点，欧洲组织就需要务实，使用市场上可用且负担得起的方案。

您曾与德国联邦数字化与国家现代化部合作开发主权AI平台。该项目提供了哪些经验？
莱因哈特：一个重要经验是：你试图解决的问题才是起点，而非为了主权而主权。
在该案例中，重点是改进重大基础设施项目的规划与审批流程。我们能够构建一个满足这些需求的解决方案，同时使用高度主权的组件，包括欧洲的AI替代方案。
这证明了一个重要观点：欧洲技术已能在许多场景中交付有意义的成果；我们只需了解它们最适合何处、能在何处创造价值。

在制定AI战略时，政府应做出哪些改变？
莱因哈特：政府需要重新思考如何采购和部署技术。
历史上，公共部门技术项目围绕漫长的规划周期、详细规格及设计用数十年的系统展开。这种方法对AI并不适用，因为技术迭代过快。
相反，政府需要更短的开发周期、更小的项目，以及通过将AI模型与已部署系统解耦来持续评估选项的意愿。AI模型、主权要求及可用技术变化太快，传统采购方式已无法应对。
这同样需要文化转变。公共机构的设计旨在最小化风险，但AI需要更多实验。坦白说，有时你必须承担更多风险。一味求稳有时无法带来切实收益。我们需要作为社会开始改变思维，奖励那些创新的人。

编者注： 本访谈经过编辑，以确保清晰与简洁。

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
Strategies for governments to more effectively pursue AI sovereignty.
A European Commission-backed e-government benchmark report warns that EU member states need to pick up the pace on digital transformation.
The annual report, released last week, identified AI sovereignty as one of Europe’s defining technology priorities.
Included in the benchmark system for the first time, AI sovereignty reflects a shift in how governments think about technological control. Governments are looking beyond where data is held to how models and infrastructure are actually built, and how they can maintain oversight of their use.
In this Q&A, Marc Reinhardt, executive vice president and public sector leader at France-based multinational IT consulting firm Capgemini, and a contributor to the report, discusses the growing importance of AI sovereignty and why governments may need to rethink traditional approaches to technology procurement and deployment in the age of AI.
Why do you think sovereignty was included in this year’s report?
Marc Reinhardt: It’s definitely a new phase for the report. The fact that sovereignty appears in the benchmark at all is significant. Historically, discussions around sovereignty were focused on infrastructure; cloud providers, data hosting and where critical systems were located. Now the debate is broadening as the EU moves to define what digital sovereignty means in practice through new policy initiatives and requirements.
In this context, it’s natural that governments are beginning to assess where they stand. They’re asking what sovereignty means across the entire digital stack, and what level of control they need over the technologies that underpin public services.
Where does AI fit within the broader digital sovereignty discussion?
Reinhardt: AI adds a completely new dimension to the debate. AI systems don't exist in isolation; they run on infrastructure and models, and they embed assumptions, values and ways of reasoning.
Countries are increasingly asking not only where their data is hosted, but also what models they are using, who developed them and whether they understand how those systems make decisions, based on the data they’ve been trained with. For some use cases, those questions may not matter very much. For others, such as welfare decisions or highly sensitive government services, they become much more important.
Compared with cloud sovereignty, which is relatively mature, AI sovereignty is still evolving, and its definition remains more fluid. But it's rapidly moving to the forefront of policy discussions.
What does AI sovereignty look like in practice?
Reinhardt: Debates around sovereignty were typically politically motivated, and demands were sometimes very abstract. Now, we’re doing more to clarify and demystify the concept of AI sovereignty.
The starting point is understanding what you're trying to protect and why. Are you concerned about data access? Resilience? Regulatory compliance? National security? Once you understand the objective, it becomes much easier to identify the right solution.
The reality is that sovereignty isn't a single solution. There’s a broad range of options available, from different cloud environments to sovereign AI platforms and locally controlled deployments. The challenge today is no longer a lack of options, but about understanding which option makes sense for which use case.
What is appropriate for publicly available business registration data may be very different from what is required for health records or other sensitive information. The level of sovereignty should reflect the sensitivity of the workload and the risks involved.
Is there a risk that pursuing sovereignty could come at the expense of AI innovation?
Reinhardt: There are certainly trade-offs, but the emerging consensus is that this isn't a binary choice.
Just as organizations increasingly operate in a multi-cloud environment, we're likely moving toward a multi-AI world where different models are used for different purposes. Some applications may use leading commercial models, while others may rely on open source or locally hosted systems where greater control is required.
The goal is really to find a balance between sovereignty and solving the problem at hand. At the end of the day, we have to solve problems. We have to find solutions to improve society and civilians’ lives. If the solutions currently available in Europe do not do that, European organizations will need to be pragmatic and use what is available and affordable in the market.
You've worked on a sovereign AI platform with the German Federal Ministry of Digitalization and State Modernization. What lessons does that project offer?
Reinhardt: One important lesson is that the problem you’re trying to solve should be the starting point, not sovereignty for the sake of sovereignty.
In this case, the focus was on improving planning and approval processes for major infrastructure projects. We were able to build a solution that met those requirements while also using highly sovereign components, including European AI alternatives.
That demonstrates an important point: European technologies are already capable of delivering meaningful results in many scenarios; we just need to understand where they fit best and where they can create value.
What should governments be doing differently as they develop AI strategies?
Reinhardt: Governments need to rethink how they procure and deploy technology.
Historically, public-sector technology projects were designed around long planning cycles, detailed specifications and systems intended to last decades. That approach doesn't work well for AI because the technology evolves so quickly.
Instead, governments need shorter development cycles, smaller projects and a willingness to continuously reassess their options by decoupling AI models from the systems they’ve deployed in. AI models, sovereignty requirements and available technologies are changing far too rapidly for traditional procurement approaches.
That requires a cultural shift as well. Public institutions are designed to minimize risk, but AI demands more experimentation. You have to sometimes, frankly, take a bit more risk. Sometimes playing it safe doesn’t translate into tangible benefits. We need to start thinking differently as a society and reward people for being innovative.
Editor’s note: This interview has been edited for clarity and conciseness.