Summary

• UMAI is available to initial pilot users within Finland's Ministry for Foreign Affairs.
• The controlled system supports material classified at national security level IV and certain other non-disclosable information.
• The ministry is working with Valtori and CGI, with lessons intended to inform a shared government AI platform.

Finland’s Ministry for Foreign Affairs has begun piloting an AI assistant capable of processing certain classified and confidential information, moving generative AI into administrative work that ordinary public services cannot safely accommodate.

The system, called UMAI, is being used by an initial group of employees in a secure environment designed for information at Finland’s national security classification level IV. It can also support work involving specified non-disclosable information, including certain personal data.

Announced on 13 October, the project involves the Government ICT Centre Valtori and technology partner CGI. The ministry will use experience gained from the pilot to help develop a common secure AI platform for central government, although that wider system has not yet been deployed across public administration.

Foreign ministries routinely handle information that cannot be entered into conventional external AI services. Diplomatic correspondence, personal records and sensitive administrative material may all be subject to restrictions governing where information is processed and who can access it.

UMAI has been developed to expand the range of work in which authorised officials can use AI assistance without removing those information handling requirements. The system operates within a controlled environment, although the ministry has not published a complete technical account of the architecture or its security assessments.

The initial phase is intended to test practical applications and gather feedback. It is not an announcement that autonomous systems will assume responsibility for diplomatic policy or statutory administrative decisions.

Testing AI in sensitive administrative work

Participating ministry units are testing a general-purpose chat assistant alongside systems tailored to particular tasks. Applications include supporting the preparation of human rights complaint matters, searching migration guidance, translation and the handling of requests for administrative review.

These processes involve significant amounts of documentation, and employees may need to locate details across several sources before preparing advice or correspondence. AI assistance could reduce some retrieval and summarisation work, but the outputs must still be checked against authoritative records and applicable rules.

Inaccurate summaries are a concern even within a secure environment. A system may omit relevant information, misinterpret a document or generate an apparently plausible statement that is not supported by its source material.

The ministry has not published measured improvements in productivity or accuracy arising from the UMAI pilot. Nor has it demonstrated that every proposed application can be performed reliably without substantial human verification.

Use in human rights and migration-related administration makes those controls particularly important because the information may concern individuals and circumstances with legal consequences. Officials need to retain responsibility for interpreting evidence and ensuring that administrative processes follow established requirements.

The planned evaluation must therefore consider more than the security of the underlying computing environment. Questions include whether employees can trace generated information to its source, how inaccurate results are corrected and whether system use creates additional documentation or oversight requirements.

Future applications may extend to reporting from diplomatic missions, situation analysis and work involving international political or economic relations. These uses remain prospective rather than confirmed functions operating throughout Finland’s foreign service.

Government control without a separate system for every department

Finland is using the project to build experience that could support shared infrastructure across central government. The Foreign Ministry is acting as a pilot organisation alongside Valtori, with the broader objective of establishing secure AI services that can accommodate public administration requirements.

Common infrastructure could reduce the need for individual agencies to develop separate systems for model access, security controls and information retrieval. It may also simplify updates and support arrangements if a suitable governance model can be established.

However, departments have different responsibilities and classification requirements. A platform suitable for one group of officials cannot automatically grant another group access to the same information, even when the technology is operated within a common government environment.

Access management, logging, retention and controls over movement of information between agencies will therefore influence the architecture. So will the division of responsibility between the government organisations and commercial suppliers involved.

The ministry connects UMAI to wider European aims concerning technological control and strategic autonomy. In practical terms, those objectives require government to understand and govern the systems processing its information, rather than assuming that domestically hosted AI removes every international technology dependency.

Procurement and maintenance costs will also affect wider deployment. AI platforms need continuing investment in computing, software updates, testing and support, while changes in models or organisational requirements can require further assurance work.

Employee experience may become a determining factor. A system with strong restrictions may still fail to attract sustained use if documents are difficult to retrieve, results are unreliable or manual checking consumes the time it was intended to save.

Conversely, a controlled service that performs defined tasks reliably could support public organisations where ordinary commercial AI tools cannot process the information required.

UMAI has reached the stage of a limited operational pilot. The ministry now intends to refine it using user feedback, with wider government ambitions dependent on what the initial deployment demonstrates about security, accuracy and practical utility.