Croatia’s ChatGPT pledge shows state AI is easy to announce

Zagreb’s latest technology pledge has a familiar shape: a ChatGPT, but Croatian. The Croatian government plans to develop a Croatian ChatGPT, and the same announcement points to state AI virtual assistants. That puts Croatia beside other governments in Europe and Eurasia that are trying to turn artificial intelligence from an imported service into a national instrument.

A national model is a procurement problem

A national ChatGPT is not a language pack. It requires decisions about who trains it, on what corpus, on whose hardware, under what security rules, and who maintains it after launch. Those choices turn a political announcement into budgets, contracts and staffing. So far, the public statements describe direction more than delivery.

Croatia’s pledge also includes state AI virtual assistants. That raises a different question: whether citizens will meet the state through a general-purpose model or through narrower assistants tied to specific services. The difference matters because a narrow assistant can be tested against error rates, privacy rules and a defined workflow; a national chatbot invites open-ended use.

A virtual assistant can be useful without becoming a national champion. It can answer questions about forms, opening hours or eligibility, then hand off to a human when confidence falls. But that only works if the underlying rules are current, the data is clean and the system is monitored after release. A state assistant that is wrong once about a benefit or a deadline can cost more trust than it saves.

Must is not a contract.

Turkey, Serbia and Russia add political volume

The regional picture is built from separate assertions. In Turkey, academics are reported to consider the Council of Higher Education’s call for AI education to be critical in academia. The reported number of artificial-intelligence initiatives in Turkey has surpassed 500. That figure is easy to repeat and hard to audit unless the count defines an initiative and tracks whether projects survive the pilot stage.

The word initiative can cover a research group, a student club, a pilot, a funded product or a dormant website. Without a definition, a total above 500 says little about scale, spending or survival. It may show enthusiasm. It does not show capacity.

Serbia’s position is framed as necessity: it must develop its own artificial-intelligence tools. Must is not a contract. It signals pressure, but not money, ownership or users.

Russia’s announcement is more specific in one way: schoolchildren will begin to be taught how to use artificial intelligence. That is a curriculum decision, not a model launch. It can create demand for tools, teacher training and safeguards, but the announcement stops at the intention.

Sovereignty becomes a market when someone pays

For vendors, national AI programmes can look like a new procurement category: language models, cloud capacity, integration, evaluation, hosting and support. For governments, the risk is fragmentation. Each country may demand separate models, separate data rules and separate maintenance, raising costs and duplicating work.

Procurement is where sovereignty meets arithmetic. A model must be hosted somewhere, updated by someone and evaluated against failure. If each state insists on its own stack, vendors must support many small deployments rather than one large one. If states share infrastructure, they must agree on data, language and accountability. Neither path is automatic.

The announcements do not show that this market has arrived. They show that officials are trying to create it. The test is mundane: budgets, tenders, service-level agreements, liability rules and evidence that citizens use the tools. Until those appear, Croatia’s ChatGPT, Turkey’s 500 initiatives, Serbia’s necessity and Russia’s classroom plan are markers of intent. They are not proof of production.