Szentgáli‑Tóth Boldizsár, a lawyer at K‑Monitor, sets out six practical proposals intended to boost transparency and accountability around government use of artificial intelligence (AI). The piece argues that while transparency has been a central theme in recent Hungarian public debates — for example in the anti‑corruption legislative package approved by the National Assembly in June 2026 — there has been little focus on how algorithms and AI tools used by public bodies actually operate.
The central problem
Public records and procurement databases can show how much was spent on developments, but they typically do not reveal what an algorithm does, which data it uses, or what decision logic it follows. Meanwhile, public bodies increasingly deploy systems that pre‑screen cases, prioritise inspections, or analyse large camera networks.
The article cites concrete examples: the National Tax and Customs Administration (Nemzeti Adó‑ és Vámhivatal, NAV) has for years used machine‑learning models that analyse invoice, cash register and transaction data as well as social media content to select targets for inspection. The police’s Szitakötő system connects tens of thousands of cameras and is capable of face recognition. For neither system is there a public, detailed description of the data inputs, logic, or human oversight involved.
K‑Monitor’s starting point is simple: what is invisible cannot be questioned, and what cannot be questioned will sooner or later be abused. The timing is favourable for national action because the EU’s 2024 artificial‑intelligence regulation (AI Act) has been complemented by an Omnibus package that delayed creation of an EU database for the highest‑risk AI systems until 2028; K‑Monitor argues Hungary should not wait for that timetable.
The six proposals
-
A public, searchable registry of government AI: every ministry and subsidiary institution should have 90 days to report which decision‑making or decision‑supporting AI systems they operate or plan to operate. The registry would show, per system, who uses it, for what purposes, who is affected, what data it processes, and where complaints can be lodged. Disclosure would be monitored and enforceable by a competent authority such as the National Media and Infocommunications Authority (Nemzeti Média‑ és Hírközlési Hatóság) or the Nemzeti Adatvédelmi és Információszabadság Hatóság (NAIH). The proposal notes international precedents: national registries in the Netherlands, a city registry in Helsinki, and many state developments in Estonia being open source.
-
Right to human review when AI influences a decision: current practice offers no uniform path for review when someone loses a benefit or becomes the subject of an inspection because of an algorithm. The EU AI Act grants a right to explanation in some cases, but national procedure is inconsistent. K‑Monitor recommends amending Hungary’s general administrative procedure act (Ákr.) so that, when AI played any role, the authority must state this in writing, explain the main considerations behind the decision and how to request human review. The proposal also suggests empowering the Commissioner for Fundamental Rights (alapvető jogok biztosa) to investigate AI‑related complaints and establishing a single‑window complaint forum.
-
Transparency of the national AI policy body: the Hungarian Artificial Intelligence Council (Magyar Mesterséges Intelligencia Tanács), established in 2025, sets the strategic direction for national AI policy but has disclosed little about its composition or procedures. Information available is limited to a short press release announcing the founding meeting, a single Facebook post about the March 2026 session, and a non‑public rulebook. K‑Monitor proposes that law require the council to publish its rules of procedure, up‑to‑date membership and minutes or summaries of its meetings, and allow civil society at least biannual opportunities to comment on its work.
-
Reveal who sells AI to the state: most government AI systems are procured rather than built in‑house, but procurement records rarely indicate whether an acquisition concerns AI. Amending the public procurement law (Kbt.) could make it mandatory to state whether AI is involved and, if so, to classify the system by its risk level under the EU AI Act. Parties involved in development or evaluation should declare any conflicts of interest. For bespoke developments, suppliers should hand over source code and full technical documentation to the state and commit to their potential publication so that systems developed with public funds remain public assets rather than closed, unaccountable products of private vendors.
-
Treat AI‑based decisions like other administrative acts under freedom‑of‑information rules: Hungary’s freedom of information law (Infotv.) defines public interest data broadly, which could cover information about AI systems. In practice the law does not name AI explicitly and many bodies do not record or acknowledge their use of such systems, meaning there is nothing to disclose. K‑Monitor suggests the law be amended to explicitly name AI and require that decisions made or prepared with AI be documented to the same standard as any other administrative act, with key data proactively published and all records linked to the registry proposed in point 1.
-
Internal rules to address ‘shadow AI’: many civil servants already use chatbots and other external AI tools without their employer’s knowledge — a phenomenon often called ‘‘shadow AI’’. Copy‑pasting documents or uploading spreadsheets to free online services can expose personal data, trade secrets or other sensitive information to external providers. K‑Monitor recommends each public body adopt an internal AI policy that specifies which tools are permitted, provides secure alternatives, bans certain data from being uploaded to external services, designates a responsible official for decisions supported by AI, and mandates training for staff.
What can citizens do now?
The author notes citizens do not have to wait for legislative changes to act: when an authority issues an unclear decision, people may ask what criteria and tools led to it and exercise their right to inspect case files. Anyone can submit a freedom‑of‑information request to ask whether a public body uses AI and for what purpose; the KiMitTud platform can help by making questions and official answers public. If personal data have been handled unlawfully, complaints can be filed with the Nemzeti Adatvédelmi és Információszabadság Hatóság (NAIH).
Conclusion
The six proposals are interlinked: a registry would reveal where AI is used by the state; effective remedies and human review would allow those affected to challenge adverse algorithmic decisions; public scrutiny of the Hungarian Artificial Intelligence Council would show who shapes policy; procurement transparency would uncover vendors and risks; freedom‑of‑information coverage would enable practical oversight; and internal policies would curb unmonitored uses. K‑Monitor recommends prioritising the registry and the right to human review as immediate measures, arguing that none of the proposals is costly or innovation‑stifling — rather, they would ensure that citizens know who decides about their affairs and on what basis, and that AI can serve a fairer, faster and more transparent public administration only if its operation is visible and accountable.
