How – Inside the AI OS: running governed agents on your own cluster, liveRegister →
INSTITUTIONS, BODIES, OFFICES AND AGENCIES OF THE UNION

Sovereign AI for EU Institutions

Twenty-four languages, twenty-seven member states, and a mandate to lead on the very rules it has to obey. The Union's institutions need AI that is European-native, multilingual by design, and governed to the standard the EU AI Act sets for everyone else.

Multilingual by design

Retrieval, agents and assistants work across the official languages on the institution's own documents, with citations to the source text.

The AI Act applied to itself

Model inventory, logging, human oversight and documentation built in, so the institution meets the obligations it asks of others.

European-native, no dependency

Built in Europe, run on European infrastructure, with open-weight models. Strategic autonomy as an operating condition.

IN PLAIN TERMS

An institution, body or agency of the Union works in every official language, across member-state administrations, on data that ranges from public to EU-classified. It has to apply the AI Act to itself, answer to the European Data Protection Supervisor and the Court of Auditors, and avoid strategic dependence on a non-European provider. The platform gives it multilingual AI on its own infrastructure, with the transparency and audit the Union asks of others.

Read this if you're leading digital transformation, IT, data, security or data protection at an EU institution, body, office or agency, or in a Commission directorate or an executive agency.


WHAT THE PLATFORM DOES HERE

For the institutions, bodies and agencies of the European Union, Scrydon is the European-native sovereign AI and data platform that runs multilingual models, agents and analytics on the Union's own infrastructure, including EU-classified networks, with policy, identity and audit built to the standard of the EU AI Act and the supervision of the European Data Protection Supervisor.

It runs in the institution's own data centres, in a European sovereign cloud or fully disconnected for classified information, with open-weight models served locally and no dependency on a non-European provider in the loop.

THE INSTITUTION'S PROBLEM

Every language, every member state, and the rules you wrote

An institution, body or agency of the Union works in twenty-four official languages, across twenty-seven member-state administrations, on data that ranges from open to EU-classified. Legislation, consultations, correspondence and case files exist in every language, so translation is a workflow rather than a feature; a policy question needs data from national administrations that are separate controllers with their own legal bases; and classified and public information sit on networks that must never be connected.

The Union also holds itself to the standard it sets for everyone else. The AI Act applies to its institutions, the data-protection regulation for the institutions is supervised by the European Data Protection Supervisor, the Court of Auditors asks how public money was spent, and strategic dependence on a non-European provider for something as central as AI is a finding in itself. The AI that helps has to be multilingual by design, European-native, and governed in a way the institution can show.

  • Twenty-four languagesLegislation, consultations, correspondence and case files in every official language; translation is a workflow, not a feature.

  • Data across member statesPolicy questions need data from national administrations that are separate controllers with their own legal bases.

  • Classified and public side by sideFrom open data to EU-classified information, on networks that must not be connected.

  • Held to the highest standardThe AI Act, the EU data-protection regulation for the institutions and the Court of Auditors all apply, and a non-European dependency is a strategic finding.

ON THE PLATFORM

Multilingual, federated, governed

The platform grounds retrieval and agents in the institution's own documents, across the official languages, with citations to the source text, so an answer about a directive, a case or a consultation can be checked in the language it was written in. Agents prepare case files, translations, impact assessments and reports, and route anything discretionary to an official, who decides; the decision and the reasoning are recorded together.

Where a policy question needs member-state data, a data space lets each national administration publish governed data products under its own access policy, and the question runs as a query across them under a named legal basis, logged, with aggregation or pseudonymisation enforced where the law requires. Around all of it, AI governance keeps the model inventory, applies policy, logs every action and records human oversight, which is the evidence the AI Act, the Supervisor and the Court of Auditors ask for.

  1. 1

    Ground

    Retrieval and agents answer from the institution's own documents, in the official languages, with citations to the source text.

  2. 2

    Federate

    Data spaces let member-state administrations publish governed data products for a policy question under a named legal basis, without a central copy.

  3. 3

    Automate

    Agents prepare case files, translations, impact assessments and reports; anything discretionary goes to an official.

  4. 4

    Govern and evidence

    Model inventory, policy decisions, human oversight and an immutable audit log produce the AI Act and supervisory evidence.

WHY SOVEREIGN

Strategic autonomy, applied to the Union's own tools

Strategic autonomy is applied to the Union's own tools. The platform is European-native and runs in the institution's own data centres, in a European sovereign cloud or fully air-gapped for EU-classified information, with open-weight models served locally. There is no inference call, licence check or update path outside the perimeter, and an external model can be enabled only by explicit opt-in under policy, never by default.

Control follows the data and the actor. Classification markings are metadata enforced at retrieval, so nothing is served above the reader's clearance even inside the same network; agents act under scoped identities and every action is attributed; a cross-administration query carries its legal basis. The audit log, the policy decisions and the model inventory are the EU AI Act and ISO 27001 evidence, produced by running the platform, and the same governance model serves the open side and the classified side, so there is one stack to accredit.

  • European infrastructureRuns in the institution's data centres or a European sovereign cloud, disconnected for classified information, with open-weight models served locally.

  • No non-European provider in the loopNo inference call, licence check or update path outside the perimeter. External models only by explicit opt-in.

  • Clearance at retrievalClassification markings are metadata enforced when data is read, so nothing is served above the reader's clearance.

  • Evidence for the supervisorAudit log, policy decisions and model inventory are the evidence for the AI Act, the Supervisor and the Court of Auditors.

Newsletter

Keep an eye on this space

What changed for your sector in European AI sovereignty, and what we learned in the field. A few times a year, no drip campaign.

A few times a year. No drip campaign, unsubscribe in one click. Privacy policy

Use cases

Use cases for EU institutions

Multilingual services, cross-administration analytics and governed AI for the Union.

Citizen-Developer Case Automation

Digital Services

Challenge

Departments have a backlog of small, repetitive workflows — a permit reminder here, an intake form there — but IT teams cannot build a bespoke agent for every one.

Solution

A governed no-code builder lets policy and operations staff assemble their own AI agents from approved building blocks, with every workflow still subject to the same access controls and audit trail as an engineered agent.

Read more

Dozens of small workflows automated by the teams that own them, without adding to the central engineering backlog or bypassing governance.

Shadow AI Discovery for EU AI Act Readiness

Compliance & Risk

Challenge

Individual teams have quietly adopted their own AI tools and agents, so the organisation cannot produce a complete inventory ahead of EU AI Act high-risk obligations.

Solution

Continuous discovery surfaces every agent and AI tool in use across departments, classifying each against EU AI Act risk categories and routing unsanctioned use into governed alternatives.

Read more

A complete, defensible AI inventory and a measurable reduction in unauthorised tool use ahead of the 2 December 2027 enforcement deadline.

Multilingual Citizen Service Assistant for Municipal Counters

Citizen Services

Challenge

Municipal counters answer the same questions about permits, registration, waste and benefits in several languages, and the answers depend on local regulations that change every council term.

Solution

An assistant grounded in the municipality's own regulations, procedures and forms answers residents and counter staff in their language, cites the article it relied on, and hands over to a civil servant for anything that needs a decision.

Read more

Shorter queues, consistent answers across counters and channels, and a record of which regulation every answer was based on.

Cross-Ministry Data Space for Policy Analytics

Policy & Statistics

Challenge

Answering a policy question about, say, energy poverty needs data from tax, social security, housing and the energy regulator. Each ministry is the legal controller of its own data and cannot simply copy it into a shared warehouse.

Solution

A data space lets each ministry publish governed products from its own systems, under its own access policy, so analysts can join them for a specific, authorised question without any ministry giving up control of its data.

Read more

Policy questions answered in weeks instead of legislative cycles, with every access logged against the legal basis that allowed it.

FAQ

Frequently asked questions

Does the platform work across all official languages of the Union?+
Yes. Retrieval and agents work on the institution's own documents across the official languages and cite the source text, and the platform is model-agnostic, so the strongest open-weight multilingual models can be served locally.
How does the platform help an institution apply the AI Act to itself?+
With a model inventory, shadow AI discovery, policy enforcement, logging, human oversight and documentation built into the way it runs. The EU AI Act page maps the controls.
Can it run on EU-classified networks?+
Yes. The complete platform runs air-gapped with open-weight models served locally, updates through the accredited channel and classification enforced at retrieval.
How do we work with member-state data without a central copy?+
Through data spaces: each national administration publishes governed data products under its own access policy, and a policy question is a query across them under a named legal basis, logged.
Is there any dependency on a non-European provider?+
No. The platform is European-native and runs on European infrastructure with open-weight models; an external model can be enabled only by explicit opt-in under policy. See sovereignty.
Do you partner on tenders and framework contracts?+
Yes, always. We are always looking to partner with primes, integrators and consortia on tenders, RFPs, framework contracts and European programmes, as the sovereign AI and data platform inside a larger bid or as a specialist subcontractor. If you are preparing a bid, talk to us early.

Or write to us

Tell us what you are working on and who should reply. A person reads it and replies within one business day.

We only use these details to reply to you. Privacy policy

Prefer to write? Email hello [at] scrydon.com and we will get back to you.

NEXT STEPS

Keep going

Three routes from here: the rules you will be measured against, the platform these outcomes run on, and the sessions where we walk through them.

Next webinar

How – Inside the AI OS: running governed agents on your own cluster, live

17 Sept 2026, 09:00

Part 2 of the Sovereign AI series, for engineers and architects. No slides after minute five: an agent gets an identity and scoped permissions, calls tools over governed MCP, retrieves from the ontology rather than raw tables, runs inside the sandbox and is stopped when it steps outside policy, and everything lands in the audit trail — then the same stack brought up on a disconnected network. Properties and refusals, shown rather than claimed.

Partners

Building the future of Data & AI together with leading innovators. Learn more.
Delaware logo