Government priority · 28.05.2026.
PPPA working document. The document is structured in two levels — Strategic Goals (what AI deployment delivers for society) and Action Plan (how it is implemented). All core principles of the EU AI Act are observed.
Time remaining until the strategy approval deadline (05.09.2026.) · Government Declaration ↗
Part A — Strategic Goals
The goals serve society — not bureaucracy. AI is a tool, not a goal in itself.
State and municipal institutions provide public services to citizens and businesses on their own initiative, without waiting for an application to be submitted. Benefits, permits and notifications are provided automatically — citizens and businesses no longer need to request or prove facts that public administration already holds.
State and municipal institutions use AI in preparing decisions so that decisions are prepared faster, are data-driven, and carry a full audit trail. Decision-makers work with AI-prepared analysis. The administrative burden on citizens and businesses is reduced.
New capacity in education, healthcare, safety and entrepreneurship. The purpose of using AI is not better bureaucracy, but a freer and more capable society with new horizons of opportunity. People add value — AI provides support.
Latvia builds an open identity and mandate infrastructure for AI agents, in which any Latvian or foreign natural or legal person can create their own "AI Twin" — a personal AI representative that acts strictly within the mandate granted to it and maintains full traceability. For individuals, this saves time in dealings with public institutions; for businesses, it reduces administrative burden and opens a digital route to the EU single market from Latvia. For the state, the "AI Twin" serves as an export product and an investment magnet — Latvia becomes a jurisdiction from which AI agents can operate reliably across Europe and the world.
Part B — Action Plan
What institutions do
Each public administration institution acts on its own, without waiting for a centralised platform or a single vendor's all-encompassing solution.
What AI does
AI supports decision-making with meaningful data analysis and a fully auditable record. AI does not merely automate — it genuinely supports decision-makers.
Risks
AI deployment carries challenges and risks that must be minimised wherever possible, without slowing the pace of implementation. The State Audit Office's 2025 audit found that 17% of public administration institutions already use AI solutions, 22% plan to introduce them, while 55% of institutions have no clear plan for AI use — underscoring the need for a concrete action plan, not merely a declarative one.
Accountability
Responsibility for a public administration institution's decisions always remains with a person — an official, in accordance with applicable law.
Implementation sequencing
The action plan's items are grouped into three implementation waves by real complexity and interdependency, not by numbering order. Wave 1 (by 31.12.2026.) covers decisions, first-round measurements, and expansion of existing mechanisms that do not require major technical development. Wave 2 (by 30.06.2027.) covers core infrastructure and standards development. Wave 3 (by 31.12.2027.) covers more complex, cross-ministerial items, or those dependent on the outcomes of earlier waves.
At present, each institution develops its own digital identifiers and metadata standards for documents, resulting in incompatible systems that cannot recognise one another. Latvia must introduce a unified national digital document identity system — one shared standard for all institutions, rather than each institution's separate solution — so that documents become machine-readable and AI can find, trace, classify, link and analyse them at the level of a single institution, a sector, and the state as a whole.
Responsible institution: to be determined by the working group
Related to: Records-as-Code · Cabinet Regulation No. 282
Public sector data is built as a national asset — structured and made available for AI development under DVI (Data State Inspectorate) oversight. A national catalogue enables reusable Latvian AI components across institutions. This is a different matter from the document identity standard in point 1 — identity determines how a document is recognised and traced; the catalogue determines which data and datasets are available for AI training and reuse.
Responsible institution: to be determined by the working group
Example: Estonia's "kratijupid"
The Latvian language in the AI economy is not merely a cultural matter — it is market-access, public-service, and national-security infrastructure. Latvia must build a Latvian-language AI dataset with legally cleared corpora, public-sector terminology and structured dictionaries, speech and dialect datasets, and an independent Latvian and Latgalian language performance benchmark.
Responsible institution: to be determined by the working group
Related to: the AI Centre's activity priorities in language technology.
All public procurement contracts exceeding the thresholds set in Directive 2014/24/EU are prepared using a machine-readable standard. Contracts become structured data that can be used for AI-supported monitoring of contract conclusion and performance.
Responsible institution: to be determined by the working group
Related to: Agreement-as-Code · IUB · Directive 2014/24/EU
A state-funded software solution is a public asset, not the property of a single institution. At present, each institution procuring an AI solution often pays to develop it from scratch, because code already developed and paid for elsewhere is not available for reuse — creating parallel, unnecessary costs across the public sector. The strategy must establish a requirement: source code for publicly funded AI solutions — except components containing sensitive, classified, or security-related information — must be published in an open, state-maintained repository under a clear open-source licence, and other institutions must be permitted to freely adapt and use it. The requirement must be built into standard procurement terms, so it does not depend on any single supplier's or project's goodwill.
Responsible institution: to be determined by the working group
Related to: point 2 (national data catalogue) · EC Open Source Strategy · industry proposal "Open-source priority in state AI systems"
Building a full AI infrastructure stack — from chips to frontier models — is not a realistic goal for Latvia; the approach must be selective. Sensitive data processing, critical infrastructure, defence, and Latvian-language services need national or closely controlled European capacity, with a continuity plan for supplier loss. Broader compute should draw on shared European resources — first and foremost AIFA-LAT (Latvia's AI Factory Antenna), which provides access to EuroHPC infrastructure without Latvia needing to fund its own hyperscale supercomputer. Global platforms remain available for commercial, lower-risk uses, with clear data classification and supplier-replacement plans. The goal is guaranteed access, not ownership of all compute capacity.
Responsible institution: to be determined by the working group
Related to: AIFA-LAT ↗ · EC Tech Sovereignty Initiative; EU Apply AI Strategy (October 2025); PPPA position "Foreign AI Compute, Local Control" ↗
Latvia publicly demonstrates a working "AI Twin" — a personal AI representative available to any Latvian or foreign natural or legal person. No new legislation is required — at level L0 the "AI Twin" prepares a document which the person reviews and signs with their own eID; at level L1 the person issues a machine-readable, e-signed, limited and revocable mandate, and the "AI Twin" acts within its bounds — every action cryptographically bound to the person with full traceability. The demonstration proves that in Latvia, people can safely entrust day-to-day interactions with the state and local government to their "AI Twin" — it serves as the foundation for a future exportable AI agent infrastructure.
The proposed model deliberately differs from Estonia's initiative, announced in June 2026, to issue AI agents separate ID codes for each function or task. Latvia should offer a simpler, more transparent principle for citizens: each citizen has one single "AI Twin" with its own verifiable identity, operating under the same rights — granted by, and revocable at any time by, the person — rather than several functionally separate agents each with a different mandate. This is not an attempt to catch up with Estonia, but a deliberately different, simpler architecture.
Responsible institution: to be determined by the working group
Related to: AI LV Exchange; AI Twin LV concept. For comparison: Estonia's AI agent ID initiative (announcement, 19.06.2026.) ↗
Each institution implements at least one pilot project comprising a working, algorithmised, auditable operational process that uses AI to support decision-making.
Responsible institution: to be determined by the working group
Related to: Process-as-Code · Annex III of the EU AI Act
Public-private cooperation is used to rapidly build AI capability in the public sector. Private partners provide technology and expertise; the public sector provides the necessary data and regulatory environment. Not procurement, but partnership with shared accountability for results.
Responsible institution: to be determined by the working group
Related to: PPPA research (2026) ↗
Expand the special regulatory environment, the "AI Sandbox" — more projects under evaluation, a faster selection cycle, and active involvement of institutions and state-owned enterprises, contributing their own data and infrastructure. Private "AI sandboxes" should also be used for pilot project needs.
Responsible institution: to be determined by the working group
Municipalities and municipal-owned companies face a fragmented market, limited resources, and high vendor-dependency risk when each municipality separately tackles the same issues with the same limited pool of suppliers, but at differing levels of competence. The strategy should provide a coordinated support mechanism — shared procurement templates with clear algorithmic transparency and risk-assessment requirements, a knowledge- and experience-sharing platform, and the possibility of pooling municipal procurement.
Responsible institution: to be determined by the working group
Example: the UK's local government AI procurement taskforce model (Ada Lovelace Institute recommendation, 2025)
All public sector decision-makers — not only IT specialists — complete structured AI governance training. The focus is on understanding the principles of AI operation in their respective field, validating AI-generated analysis, and overseeing semi-automated decisions that are fully auditable.
Responsible institution: to be determined by the working group
Example: Estonia's AI Leap 2025 ↗ · EU AI Act explainer tool (aiact.cyberfort.lv); also Finland's national AI training academy and Lithuania's AI training platform (OECD, "Building an AI-ready public workforce", 2026)
The AI literacy mandated in point 12 must be differentiated by role — board and senior management, process owners, procurement specialists, lawyers and compliance functions, data stewards and internal auditors all have different needs, and separate training programmes must be developed for them. In addition, cross-sector AI deployment support teams should be established — temporary groups placed within ministries, municipalities and SMEs that combine sector expertise, process redesign, data engineering, AI deployment, cybersecurity and change management.
Responsible institution: to be determined by the working group
Related to: point 12 (AI literacy); CSB 2025 data — 63.2% of enterprises that considered AI adoption cite lack of relevant expertise as the main barrier.
State and municipal institutions and state/municipal-owned companies should use interactive registers listing the AI systems in use, in accordance with EU AI Act requirements. Register entries are published for public information purposes where the AI system's risk class permits. The public is properly informed about the AI systems used by the public sector, and about the organisation's role and responsibility under the EU AI Act.
Responsible institution: to be determined by the working group
Related to: a private-sector initiative in this direction — a Latvian-developed interactive AI Register. International example: the UK's Algorithmic Transparency Recording Standard (ATRS) — a state-level AI system registration model.
The AI Centre established in 2025 is responsible for promoting AI technology development and coordinating the regulatory sandbox. This is a valuable step, but it addresses the AI development and innovation-promotion function — not the AI Act's market surveillance and compliance function, which remains split across VARAM, the Data State Inspectorate (DVI), the Ministry of Economics/LATAK, the Ombudsman, the Bank of Latvia, and more than ten sectoral bodies. This fragmented oversight model risks leaving businesses and citizens without a single clear point of contact for AI Act enforcement matters. The AI Centre's mandate could be expanded so that it also serves as a single point of contact for AI Act compliance matters, or such a point of contact could be established separately — but the division of responsibility among the various bodies must not remain unclear.
Responsible institution: to be determined by the working group
Related to: the AI Centre Law (adopted 06.03.2025.); VARAM's 25 February 2025 informative report; PPPA article "Who will oversee AI Act enforcement in Latvia?"
Without regular public-attitude research and citizen engagement in AI solution development, there is a risk of losing public trust even in technically successful projects — especially proactive services (Goal 1), where citizens may learn of decisions made in their interest only after the fact. The strategy should provide for a regular public trust measurement and mandatory citizen or user engagement in the development of high-impact AI projects.
Responsible institution: to be determined by the working group
Example: the UK's "Public Attitudes to Data and AI Tracker Survey" (GDS)
Every institution that pilots AI solutions on its own often ends up repeatedly tackling questions already addressed elsewhere and repeating others' mistakes, because there is no centralised mechanism that compiles which AI pilot projects delivered results, which did not, and why. This is a different matter from the data catalogue mentioned in point 2 — data is the raw material for solutions, whereas this mechanism concerns evaluating the performance of the solutions themselves after deployment. The strategy should provide for such an evaluation and experience-sharing mechanism, documenting both successes and failures and the reasons behind them.
Responsible institution: to be determined by the working group
Example: the UK's proposed "What Works Centre for AI in Public Services" (Ada Lovelace Institute, 2025)
The strategy must avoid headline indicators that measure activity rather than outcomes — the number of events, trained staff, or pilot projects may remain a secondary indicator, but not the primary one. A National AI Productivity Indicator Set must be established in four parts: the economy (AI integration into core processes, verified cost and cycle-time reduction), public administration (administrative hours freed up, processing-time reduction), sovereignty and resilience (deployability of critical systems in Latvia/EU, Latvian-language model performance), and governance (number of registered systems with a designated owner, completed risk assessments). The indicator set is launched with baseline data in 2026 and refined annually, incorporating the findings of the pilot-evaluation mechanism in point 17 as they become available.
Responsible institution: to be determined by the working group
Related to: point 16 (public trust measurement); point 17 (pilot project evaluation).
Proposals received
PPPA collects industry proposals. The summary is updated regularly.
Latvia's public administration digital transformation should move from isolated IT-project-level efforts to a unified, modular, and reusable platform architecture. The strategy should provide for a federated state digital economy and AI infrastructure, comprising an AI model usage platform, a Process-as-Code approach to algorithmising public processes, a federated data ecosystem, an API and integration registry, a reusable library of state digital solutions, and a unified security, auditability and compliance layer. Such an approach would reduce system fragmentation, integration costs and vendor dependency, while creating an AI-ready state data and process infrastructure. Implementation should begin with 2–3 MVP pilot projects in state institutions and state-owned enterprises — for example, in benefits administration, procurement analytics, or construction-process digitalisation.
Each supplier with its own proprietary format means dependency, not control. Latvia's AI strategy must introduce mandatory interoperability requirements for every AI solution in the public sector — open APIs, standardised data structures, and migration capability.
Compute capacity is a real constraint on AI development, and Latvia must address it within the context of digital, economic, climate and energy policy — not in isolation. It is advisable to pre-identify locations suitable for data centre siting (grid capacity, fibre-optic connectivity, renewable electricity, heat-supply integration, cooling) and to tie state support for data centre projects to conditions — waste-heat utilisation, local employment, reserved capacity for Latvian businesses. Success should be measured not by installed megawatts, but by the economic, resilience, and innovation value delivered per megawatt of data centre capacity.
Using real personal data for AI development creates privacy risks and legal obstacles. A national synthetic data programme would allow private sector organisations to generate and use synthetic data for the development, testing and validation of AI solutions — reducing personal data protection risks, accelerating innovation, and strengthening the international export capability of Latvia's AI ecosystem.
The Data State Inspectorate (DVI) is the principal competent authority for personal data protection and oversight of high-risk AI systems under the EU AI Act. At present, DVI dedicates only one staff member at ⅓ of full-time capacity to AI matters — disproportionate to the scope of its oversight mandate and the requirements of the EU AI Act. Without adequate funding and capacity, DVI cannot effectively carry out the oversight functions entrusted to it.
There is a risk that, without clear limits, AI systems could produce outcomes harmful to society. The strategy must establish strict competence boundaries for AI operation, and compliance must be verified through testing against extreme scenarios. The goal is to eliminate any possibility of harm.
Europe depends on non-EU suppliers for more than 80% of critical digital technologies. Latvia's AI strategy must set a clear open-source priority for public sector AI solutions — so the state retains control over the systems that make decisions on its behalf.
Related to: point 5 — "Publicly funded AI solution code — publicly available"
Summary
The PPPA working document comprises 4 strategic goals and 18 action plan items with measurable success indicators, grouped into three implementation waves. An additional 7 industry proposals have been compiled.
Proposal submissions opened on 30.05.2026. The summary will be submitted to the responsible authorities ahead of the 05.09.2026. deadline.
Last updated: 30.07.2026.
Submit a proposal
Do you have an idea for what Latvia's National AI Strategy should include? Send it to [email protected]. The best proposals will be published on this page and included in PPPA's compiled summary, which will be submitted to the responsible authorities.
Please structure your proposal as follows:
International experience
Best practice from countries that have already implemented AI strategies in the public sector.
Estonia
In 2019, Estonia became the first EU country to introduce a national AI strategy, focusing on reusable AI components in the public sector.
KrattAI (OECD.AI) ↗Estonia
Estonia's framework for experimentation by state institutions — a structured approach to rapid testing and deployment of AI in the public sector.
accelerate.ee ↗Estonia
Estonia's national programme for integrating AI into education and public administration. Mandatory AI training with a practical focus.
e-estonia.com ↗EU
The European Commission's June 2026 package of initiatives — an EU Open Source strategy, interoperability requirements, and a plan to reduce dependency.
EC document ↗Latvia
An EU- and nationally-funded project (€8.4m, 2026–2028) — a national AI competence centre in cooperation with RTU, the University of Latvia, and the LUMI consortium.
VDAA announcement ↗Latvia
The special regulatory environment ("sandbox") — Process-as-Code is one of the first 3 projects selected (May 2026).
AI Centre Sandbox ↗