European Parliament: Archibot, a Claude-powered AI archivist — 80% less time searching 2.1M documents

• The EU transforms democratic access with Claude (case study one-sheeter) — Anthropic (официальный кейс, PDF)
• An AI tool is making the European Parliament's history easier to understand in all EU languages — Euronews, 2024-10-22
• EU Parliament expands archive access with Anthropic's Claude — Electronic Specifier, 2024

Background

The European Parliament's Archives Unit preserves documents dating back to 1952 — the earliest years of the European project: resolutions, positions, policies, and records of inter-institutional negotiations. It is the institutional memory of European democracy, and it is vast: more than 2.1 million official documents.

In October 2024 the Parliament unveiled a tool that made this collection accessible conversationally: "Ask the EP Archives," a.k.a. Archibot — an AI assistant built on Anthropic's Claude, deployed in Amazon Bedrock and available right on the archives website. The announcement came on October 21, 2024 and was widely covered — including by Euronews — making the project one of the first publicly documented generative AI deployments inside European Union institutions.

The target audience matters. Archibot was built not as an internal archivists' tool but as a public service: for researchers, policymakers, educators, students, and ordinary citizens worldwide. Ludovic Delepine, Head of the Archives Unit, frames the motivation in terms unusual for an IT project: "The access to information and all the tools we provide in order to be able to make use of this information are really the foundations of a democratic project."

Anthropic itself is an AI safety and research company founded by former OpenAI executives and backed by Google and Amazon (as Euronews introduces it in its launch coverage). For Anthropic, the case became a showcase of Claude in government: co-founder and Head of Policy Jack Clark accompanied the launch with the formula that "by making vast archives instantly accessible to anyone, anywhere, Archibot sets a new standard for open governance." Beyond the rhetoric, the case is interesting for exactly what the public sector rarely offers: it is public, it has a measurable impact metric, and anyone can try it — just open the Parliament's archives website.

Problem

The archive had two chronic problems the Archives Unit itself names directly: making the collection easily searchable and ensuring accessibility for all. Traditional archive navigation was time-consuming and required specialized knowledge — of the collection's structure, terminology, and procedures. In practice this meant mostly professional researchers willing to dig could actually use it.

The second barrier was language. Historically, much of the archive was available only in French. For an institution representing dozens of countries with two dozen official languages, that is a fundamental constraint: a document a citizen cannot read in their own language is effectively closed to them, even if formally "published."

The third dimension is specifically governmental: control and data sovereignty. A public institution cannot simply "plug in a chatbot": it is accountable for where data goes and how content is used. Delepine states the requirement bluntly: "If we use generative AI, we permanently need to be under control of the solution that we've built. We didn't want to have any data that could be collected or extracted for other purposes." Add zero tolerance for unreliability: the Parliament's archive cannot afford an assistant that invents resolutions or distorts positions — researchers and journalists will cite its answers. For an institution whose legitimacy rests on transparency, limited access to its own history was not an operational inconvenience but a fundamental gap between declaration and practice.

Solution

The Parliament integrated Claude via Amazon Bedrock and built Archibot — an assistant available on the archives website. Choosing Bedrock as the deployment venue directly answers the control requirement: the model runs in a managed cloud perimeter with guarantees that data is not collected or used for extraneous purposes — precisely the condition Delepine called mandatory.

The implementation strategy recorded in Anthropic's case study had five elements. First — advanced search and summarization: the user asks in natural language, the system finds relevant documents across the collection dating to 1952 and synthesizes them into a coherent answer. Second — full multilingual support beyond French: coverage of all EU member-state languages, so a Greek and an Estonian work with the same collection, each in their own language. Third — Constitutional AI, Anthropic's approach to trustworthiness and hallucination reduction, designed into the project as the condition for trusting answers in a governmental context. Fourth — report building: Archibot not only answers questions but helps extract and organize information into structured reports for the user's task. Fifth — global accessibility: the tool is open to everyone through the Parliament's website, with no special access required.

The delivery model is also characteristic: not "a vendor shipped a box" but partnership work between the Parliament and Anthropic on scope, sources, and quality control, with ownership of the solution staying with the institution. Delepine stresses permanent control as an operating principle, not a one-time procurement checkbox.

What it looks like for the user: a researcher asks about the Parliament's position on a specific policy in the 1980s — and gets a document-grounded synthesis; a student asks for the history of inter-institutional negotiations — in their own language; a policymaker builds a report on precedents without commissioning a memo from the archivists. The conversational interface removes the skill of "knowing how to search an archive" from the equation — the very specialized-knowledge barrier that had kept the collection professional-only.

The project is not frozen at launch state: the Parliament keeps expanding Archibot's capabilities, adopting the latest Claude updates, including 3.5-generation models. For a government system that is an achievement of its own: the model update loop is built into the service lifecycle rather than requiring a new tender for every version.

Result

The case's key metric: document search and analysis time dropped by 80% while maintaining high accuracy and security standards. The system processes and provides access to more than 2.1 million documents in multiple languages. The user base is diverse: researchers and policymakers get fast access to historical context and precedents, and the case records the educational effect separately — hundreds of educators and students use Archibot as a window into the history of European parliamentarism. The history of European democracy since 1952 became available to any citizen, conversationally, in their own language.

Framing. The 80% comes from Anthropic's official case study (an interested vendor) without disclosed methodology: it is unknown which scenarios were measured and against what baseline. Absolute usage figures (queries, unique users) were not published; "hundreds of educators and students" is the only audience-scale reference. No independent audit of Archibot's answer accuracy is publicly available.

Editorial analysis. First: this case is a ready-made template for the public sector, and its strength lies not in the technology but in the choice of proving ground. A historical archive is the ideal starting point for generative AI in government: the data is already public (minimal privacy risk), an error doesn't block a public service, and the impact is public and legible to auditors and citizens alike. The contrast with typical government deployments — which start with citizens' sensitive data and end in scandal — is instructive. Second: Delepine's requirement of permanent control over the solution and its data is a formula that captures the broader European approach to AI procurement: not "buy a smart bot" but "own a governed instrument." For vendors this means that in the EU you sell not a model but the control perimeter around it. Third: multilinguality here is not a feature but the very substance of the project: the technical search barrier was removed equally for everyone, but it was the language barrier that separated a "formally open" archive from a genuinely accessible one. That is a useful lens for any "open data" project: openness is measured not by the fact of publication but by the cost of access for a specific person. Finally, the template is plainly replicable: national archives, parliamentary libraries, and municipal document collections worldwide sit on the same two problems — searchability and accessibility — and Archibot hands them a ready reference architecture: public data + a governed cloud perimeter + multilingual conversational access. The only question is who replicates it next, and whether they publish their own 80%.

Lessons learned

1. Government can adopt generative AI publicly and measurably: 80% search-time savings is a metric both auditors and citizens understand.
2. Start with archives: historical documents are a safe proving ground for AI in government (data is public, an error blocks no service) with fast public impact.
3. Multilinguality turns an archive from an internal resource into an instrument of democracy: the barrier removed is linguistic, not just technical.
4. Control is an operating principle, not a tender clause: the requirement to "permanently be under control of the solution" (Delepine) is a working formula for government AI procurement.
5. Regulated environments need a trust architecture: Constitutional AI and a governed perimeter (Bedrock) were designed in from the start, not bolted on.
6. A partnership delivery model (jointly defined scope, retrieval sources, and quality controls) works better in government than a vendor handoff.
7. Build model upgradability into the service lifecycle: Archibot moves onto newer Claude generations without rebuilding the project.

Frequently asked questions