Skip to content
Industries · Public sector

AI for the public sector: citizens' data stays inside

Updated
By
The MagnolIA team
On this page
14 min read

Short answer

A public body can use AI with the data kept inside its own network. MagnolIA installs a machine in the organisation's network to prepare drafts and reviews of case files and decisions and information for citizens, with a per-user usage log and a verifiable audit trail. It is designed with Spain's ENS security framework in mind and MagnolIA does not advertise any ENS certification.

Back to contents

Data sovereignty: what does it mean for a public body?

It means a citizen's file is processed and stored on a machine belonging to the body itself, and nobody outside receives it. With a cloud AI, every query travels with the document inside it to a provider's servers. With on-premise AI, the document goes from the civil servant's desk to the machine installed in the administration's network and back.

  • The model runs inside. These are open-weight models, loaded on the machine installed in the body's network.
  • MagnolIA does not see the content. Only the monitoring of the machine's status travels outside: load, health and security.
  • It does not depend on an external service. In the installation this site describes as a real case, the machine was tested with the internet cut off before handover and kept reading documents and answering. What works and what does not is detailed in air-gapped AI.
  • The log belongs to the body. Who used what and when stays in the installation itself, not in a third party's hands.
Back to contents

ENS, GDPR and the AI Act: what does the design cover, and what does it not?

The ENS (Esquema Nacional de Seguridad, Spain's National Security Framework) is the mandatory security standard for Spanish public-sector information systems. It is regulated by Royal Decree 311/2022, whose article 2 applies it to the entire public sector and also to the information systems of private entities that provide services or solutions to it.

MagnolIA design measures built with the ENS in mind
MeasureWhat it does
Role-based accessEach person signs in with their own user; permissions can be limited by model and by machine.
Two-factorSystem administration requires two-factor authentication.
Encrypted entry pointA single entry point with TLS. Remote access, if enabled, runs through an outbound tunnel: no port open towards the internal network.
Isolated networkCompute and internal services are not exposed to the user.
AuditA chained, verifiable log, kept for two years, that remains in the installation.
BackupsEncrypted with AES-256, on the storage the body chooses.
These are technical measures of the product. They are not a certification and do not replace the system categorisation and audit that fall to the public body.

The GDPR works in a similar way: because the document is not sent to an outside provider and does not leave the EU, the assessment is simpler, but it remains the job of the body and its data protection officer. The EU AI Act, in turn, imposes obligations according to the risk of each use; keeping the AI in your own network does not exempt you, but every use is logged. The current timetable is in the EU AI Act timeline, and the three frameworks, one by one, in AI and the GDPR, ENS and AI Act.

Back to contents

What does AI do in a public administration?

The reference deployment, described in the on-premise AI use cases, works on case files, decisions and citizen enquiries. It is used by civil servants, citizen services and management. The AI prepares; the person reviews and decides.

  • Drafts and reviews of case files and decisions. Open the file and ask for a draft or a review, which the civil servant corrects and signs.
  • Information for citizens. Draft replies to enquiries based on the body's own documentation.
  • Querying regulations and internal documents. The server generates embeddings for semantic search and RAG over your own documentation, with document collections that are switched on in corporate mode.
  • Meeting minutes. MagnolIA Nexus transcribes meetings with the speech model on the server, identifies who is speaking and leaves minutes with a summary. The audio does not leave.
  • Anonymising documents before sharing them. The Nexus anonymiser replaces names, ID numbers and other data and keeps the key to revert it on the user's computer.
  • Office documents. The models create and read Excel, Word, PowerPoint and PDF files.
Back to contents

Does it integrate with corporate sign-in and the SIEM?

Yes. Corporate mode includes sign-in with the organisation's account (SSO) and organisation-wide policy. On the server there is audit output for the SIEM, model routing, document collections and encrypted backups. Everything ships switched off until the organisation enables it.

Users only use the chat, the app or the endpoint: they have no technical access to the system, the models or the infrastructure. MagnolIA handles operation and maintenance. API keys are stored only as hashes, have a configurable expiry and are revoked instantly.

Back to contents

Which specific tasks does it take on in a council or a public agency?

The same division of labour works in a town council, a provincial authority or an agency: the AI reads and drafts, and whoever holds the legal authority reviews, decides and signs.

On-premise AI in a public body: what the AI prepares in each unit and who decides
UnitWhat the AI preparesWho decides
Case handlingThe summary of a long case file, what is missing from the documents submitted, a draft report or draft decisionThe official handling the case and the body that decides it
Citizen servicesDraft replies to enquiries, based on the body's own documentsThe person dealing with the enquiry, who checks and sends
Secretariat and committeesMeeting transcripts and draft minutes with a summaryWhoever certifies the minutes, who corrects and approves them
ProcurementReading tender documents and long PDFs: deadlines, requirements and clauses locatedThe procurement unit and the evaluation panel; the AI does not score bids
Legal service and internal rulesAnswers about the by-laws, instructions and circulars the body has loadedThe lawyer or officer, who checks against the official text
Transparency and recordsCopies with personal data replaced before a document is sharedThe person in charge, who reviews every item flagged
Every row uses functions described above: reading documents, searching the body's own documents, minutes with MagnolIA Nexus and reversible anonymisation. None of them decides anything on its own.
Back to contents

Can AI decide a case on its own?

That is not the use this page describes, and Spanish law gives the difference a name. Law 40/2015 on the Legal Regime of the Public Sector (in Spanish) defines automated administrative action as any act or action carried out entirely by electronic means by a public administration within an administrative procedure, with no direct involvement of a public employee (article 41(1)).

For those cases, article 41(2) requires the body to establish beforehand which unit is responsible for defining the specifications, programming, maintenance, supervision and quality control and, where appropriate, auditing the system and its source code, and to state which body answers for the purposes of appeal. Across the EU, Article 22 of the GDPR gives everyone the right not to be subject to a decision based solely on automated processing which produces legal effects concerning them or similarly significantly affects them, with the exceptions the article itself lists.

AI as support
The AI prepares a summary or a draft; a public employee reads it, corrects it and takes ownership. This is the use on this page.
Automated action
The system produces the act with no direct human involvement. That is a different project, with its own responsible body, legal analysis and controls.
Back to contents

What if it gets it wrong? Who answers to the citizen?

A language model can state something false with confidence: a miscalculated deadline, a detail the file does not contain, or a reference to a provision that does not exist. The AI neither signs nor holds any authority: the act still belongs to whoever issues it, just as when another person prepares the draft. How responsibility is apportioned in a given case is a legal question this page does not settle; what the body can do is organise the review.

  • Every legal reference is checked against the official text. The official gazette, not the model's answer.
  • The facts in the draft are checked against the file. Dates, amounts, identities and deadlines.
  • Answers drawn from your own documents are not infallible either. Searching the loaded by-laws reduces errors; it does not remove them.
  • Who used the AI, and what for, is recorded. In a MagnolIA installation the per-user log and the chained audit trail stay with the body and can be sent to the SIEM.
  • Failures are reported. A mailbox or a named person to tell when the AI gets something significant wrong.
Back to contents

What should a public body not hand over to AI?

Five uses of AI that a public body should avoid or treat as a separate project
UseRiskAlternative
Granting, refusing or penalising with nobody reviewing itIt may stop being support and become automated action, with its own requirementsA draft reviewed and signed by whoever holds the authority
Scoring bids or ranking applicationsThe reasoning can no longer be explained by the person who signsUse it to read and locate; the assessment belongs to the panel or the officer
Replying to citizens without saying it is a systemLoss of trust, and wrong answers nobody has seenA reply reviewed by a person, or a clear notice that a system is answering
Pasting case files into a public AI from a personal accountThe document leaves the body with no contract and no recordA tool inside the network, with a corporate user
Treating the model's answer as a source of lawNon-existent or repealed provisions in an administrative actAlways check against the official gazette
The risk column is practical caution, not a legal opinion: the legal classification of each use is for the body itself.
Back to contents

What should a public body ask an AI provider before contracting?

These seven questions work for a preliminary market consultation or for drafting the tender specification, whoever the provider is. The third column records only what MagnolIA's website already states; where nothing is published, it says so.

Seven questions for an AI provider to a public body, and what is on record for MagnolIA today
QuestionWhy it mattersWhat is on record for MagnolIA
Where does the model run, and where do documents and logs stay?It determines who processes the data and under which jurisdictionOn the machine installed in the body's network; only status monitoring travels out
Can the provider evidence conformity with the ENS?Spain's National Security Framework also covers those who provide services to the public sectorIt claims no certification or declaration of conformity; it offers design measures
Does it integrate with our corporate identity and our SIEM?Without that, joiners, leavers and audit are managed separatelySSO in corporate mode and an audit feed for the SIEM, switched off until activated
Who chooses and updates the models, and how are we told?A change of model can change the answersMagnolIA selects, tests and updates them; how and when you are told of a change is worth putting in writing
Does it work offline, and how is it updated on an isolated network?Some networks cannot have an outbound internet connectionWhat has been tested and what must be agreed is in AI without an internet connection
Who accesses the machine to maintain it, and how?Maintenance is privileged accessMagnolIA operates it; remote access is optional and runs through an authenticated outbound tunnel
What happens to the hardware and the data when the contract ends?It avoids depending on a provider to get back what is yoursAt the end you choose to keep, renew or return the machine; how the data is erased if it is returned is worth putting in writing
How an AI system fits into ENS categorisation and its security measures is covered in AI and Spain's ENS; it is only signposted here. The general framework is in AI and the GDPR, the ENS and the AI Act.
Back to contents

Where does a public body start? Steps and a checklist

  1. Pick two or three support tasks

    Case-file summaries, draft replies, minutes. Tasks where a person already reviews the result.

  2. Bring in those who need a say

    The secretariat or legal service, the data protection officer and the security lead. It is better before choosing a tool than after.

  3. Decide where the data lives

    If case files cannot leave, the model has to be inside. The comparison is in on-premise AI vs cloud AI.

  4. Write the rules of use

    What may be asked, what may not, and who reviews each output. One page is enough to start.

  5. Pilot with one unit

    With real case files inside the network and a log of errors. It is extended once the unit is satisfied.

Checklist before the first real case file

  • The model runs on a machine belonging to the body, and we know who maintains it.
  • Each employee signs in with their corporate user.
  • The legal service and the data protection officer have seen the intended uses.
  • No use produces an act without a person reviewing and signing it.
  • There is a written instruction to check legal references and facts.
  • The usage log is kept and, where appropriate, reaches the SIEM.
  • Whatever the provider cannot evidence (the ENS, for example) has been identified and assessed.
Back to contents

How is it installed, and what stays inside the body's network?

MagnolIA delivers it installed and managed. The reference deployment for a public administration starts on Núcleo L, and capacity is added without interrupting service.

  1. Assessment

    We look at how many people will use the AI, what for, and which systems it needs to talk to. That sets the capacity.

  2. Installation

    MagnolIA installs the machine in your office. You do not buy it: it is included in the fee.

  3. Tuning and integration

    The models are loaded and tuned, and the integrations with your systems are set up.

  4. Pilot

    A group of 10 to 20 people works with it on their own documents.

  5. Production

    The whole staff, with corporate sign-in, monitoring and backups.

Stays inside
Prompts, documents and answers. They are processed on the machine installed in your network and stay there.
Also inside
The per-user usage log and a chained, verifiable audit trail, kept for two years.
Leaves
Only the monitoring of the machine's status: load, health and security. MagnolIA does not see the content.
Access
Each person signs in with their own user and password through a single entry point encrypted with TLS. Remote access is optional and runs through an authenticated outbound tunnel.

The machine is hardware as a service, included in a fixed fee that does not depend on usage: see what on-premise AI costs. The basic definition is in what on-premise AI is. To see anonymisation and minutes at work on a sample document, the LAB has the online document anonymiser and meeting minutes from an audio file; it runs on MagnolIA's servers, so do not upload a real case file there.

Back to contents

Frequently asked questions from public bodies

Can a public body use AI with citizens' data?

A public body can use AI with the data kept inside its own network. MagnolIA is designed with the GDPR, the EU AI Act and the ENS (Spain's National Security Framework) in mind, and does not advertise any ENS certification. The typical use is drafting and reviewing case files and decisions and informing citizens, with per-user usage records and a verifiable audit log.

Can AI decide an administrative case on its own?

A support AI does not decide cases: it prepares a summary or a draft that a public employee reviews, corrects and signs. When an act is produced with no direct human involvement, Spain's Law 40/2015 calls it automated administrative action and requires establishing beforehand who supervises the system and who answers an appeal. That is a different project.

What can a local council use AI for?

A local council can use AI as support for reading and drafting tasks: summarising case files, preparing draft reports and replies to residents, writing minutes from a meeting, locating requirements in tender documents and querying its own by-laws. In every case a person reviews the result, and with on-premise AI the documents do not leave the council's network.

What should a public body ask an AI provider?

A public body buying AI is wise to ask the provider where the model runs and where the logs stay, whether it can evidence conformity with Spain's ENS security framework, whether it integrates with corporate identity and the SIEM, who updates the models, how the machine is maintained and what happens to the data afterwards. MagnolIA claims no ENS certification.

Is MagnolIA certified under Spain's National Security Framework (ENS)?

MagnolIA does not claim any certification under the ENS, the security framework that Spanish public bodies and their suppliers must follow. What it offers are design measures built with the ENS in mind: role-based access, two-factor authentication for administration, an encrypted entry point, an isolated network, a chained, verifiable audit log kept for two years and AES-256 encrypted backups.

Does it integrate with company single sign-on and with a SIEM?

Yes. MagnolIA's corporate mode includes sign-in with the company account (SSO) and organisation-wide policy. On the server there is audit output for the SIEM, model routing, document collections and encrypted backups. Everything comes switched off until the organisation itself enables it.

Sources

External figures on this page link to their original source.

  1. BOE (Spain's Official State Gazette). Royal Decree 311/2022 of 3 May regulating the National Security Framework, ENS (consolidated text, in Spanish). www.boe.es/buscar/act.php?id=BOE-A-2022-7191

  2. EUR-Lex. Regulation (EU) 2016/679, General Data Protection Regulation (GDPR). eur-lex.europa.eu/eli/reg/2016/679/oj

  3. BOE (Spain's Official State Gazette). Law 40/2015 of 1 October on the Legal Regime of the Public Sector, consolidated text; article 41, automated administrative action (in Spanish). www.boe.es/buscar/act.php?id=BOE-A-2015-10566

Shall we talk about your organisation's requirements?

Tell us what your procurement requires, how many people would use it and on which documents. If there is anything we cannot evidence, we will say so.

Request a pilot