Part of the AI Regulation News hub.
Hungary's energy regulator gives itself a standing right to analyse the transmission system operator's hourly learning algorithm on historical data, in decree 8/2026. (VIII. 28.) MEKH, in force 30 August 2026
The decree never says artificial intelligence. It never says machine learning. What it does say, at point 8.1.2, is that the Authority is entitled to run detailed analyses on historical data of the hourly learning algorithm that sets the transmission system operator's reserve procurement needs. That is an algorithmic audit right arriving through energy cost regulation.
Bottom line: A binding decree of the Hungarian Energy and Public Utility Regulatory Authority, dated 28 August 2026, published in Magyar Kozlony 2026/120 and in force 30 August 2026. It replaces a methodological guide that previously sat on the Authority's website.
Who this affects: The transmission system operator as the duty-holder. Point 8.1.2 names only az atviteli rendszeriranyito, that role, rather than a company. The Authority, MEKH, holds the analysis right. Practically, anyone building or operating forecasting models whose outputs feed a regulated cost base.
The provision: Point 8.1.2. The Authority is entitled to carry out detailed analyses on the basis of historical data concerning the operation of the hourly-resolution learning algorithm that performs precise demand determination for reserve procurement needs, based on forecasts of PV production, weather and system load.
What it is not: It is not an AI instrument by its own vocabulary. The phrase mesterseges intelligencia, artificial intelligence, appears zero times. Gepi tanulas, machine learning, appears zero times. The phrase used is tanulo algoritmus, learning algorithm, and it appears three times: in point 8.1.2 and in its two 8.1.3 subpoints, which concern FRR and aFRR demand determination.
Consequence: No fine attaches to point 8.1.2. The decree's mechanism is cost recognition: costs the Authority does not accept are not recognised in the price cycle. That consequence follows from the decree's architecture rather than from any sanction sentence attached to 8.1.2 itself.
Primary sources: Magyar Kozlony 2026. evi 120. szam (contains 8/2026. (VIII. 28.) MEKH rendelet)
- Instrument
- 8/2026. (VIII. 28.) MEKH rendelet az arszabalyozasi ciklus indulo arainak meghatarozasat megelozo eszkoz- es koltsegfelulvizsgalatra vonatkozo modszertanrol
- Authority
- Magyar Energetikai es Kozmu-szabalyozasi Hivatal (MEKH), the Hungarian Energy and Public Utility Regulatory Authority. Signed Dr. Juhasz Edit, elnok
- Jurisdiction
- Hungary
- Publication
- Magyar Kozlony 2026. evi 120. szam
- Date of legal event
- 28 August 2026, the decree's own date
- Entry into force
- 30 August 2026. Section 8: Ez a rendelet 2026. augusztus 30-an lep hatalyba
- Status
- Binding and in force
- Bindingness
- A rendelet, a binding decree. Section 11 replaces a methodological guide previously published on the Authority's website
- Duty-holder
- Az atviteli rendszeriranyito, the transmission system operator. Point 8.1.2 names only that role. MAVIR appears elsewhere in the document in data-form and contract references, but the operative point does not name it
- Right-holder
- A Hivatal, the Authority (MEKH)
- The algorithm provision
- Point 8.1.2, on historical-data analysis of the hourly learning algorithm used for reserve procurement demand determination
- AI terminology in the decree
- mesterseges intelligencia: 0. gepi tanulas: 0. neuralis: 0. algoritmus: 3, all three being tanulo algoritmus, in point 8.1.2 and in subpoints 8.1.3.1 and 8.1.3.2
- Sanction
- None attached to point 8.1.2. The decree's consequence mechanism is non-recognition of costs, not a fine
- Editorial Note
- Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.
- Primary source
- https://magyarkozlony.hu/hivatalos-lapok/Nlmmhf3TJXwkyoaMlhhT6a8c7a7d22c3c/dokumentumok/6eb523e0bd643b2b165248273963f11c321d7299/letoltes
What point 8.1.2 says, and why we are reporting an energy decree
The operative sentence, in the original: A Hivatal jogosult historikus adatok alapján részletes elemzéseket végezni a tartalékbeszerzési igények alapját képező, előrejelzések (PV-termelés, időjárás, rendszerterhelés) alapján pontos igénymeghatározást végző, órás bontású tanuló algoritmus működésével kapcsolatosan.
Literally: the Authority is entitled to carry out detailed analyses on the basis of historical data in connection with the operation of the hourly-resolution learning algorithm that performs precise demand determination, based on forecasts of PV production, weather and system load, forming the basis of reserve procurement needs.
Strip the energy vocabulary and the structure is familiar. A regulator has written itself a standing right to inspect how a specific model behaves, retrospectively, against the data it actually ran on. Not a right to see the code, and not a conformity assessment. A right to analyse operation using historical data.
That is an algorithmic accountability mechanism, and it did not arrive through an AI statute. It arrived through the methodology for reviewing assets and costs before the starting prices of a price regulation cycle are set.
The decree never uses AI vocabulary, and that is worth saying plainly
We checked, because the temptation to file this as an AI story is exactly the sort of thing that produces a wrong article.
Mesterseges intelligencia, artificial intelligence, appears zero times in the decree. Gepi tanulas, machine learning, appears zero times. Neuralis appears zero times. Algoritmus appears three times, and all three are tanulo algoritmus, learning algorithm: once in point 8.1.2 and once each in subpoints 8.1.3.1 and 8.1.3.2, which deal with FRR and aFRR demand determination.
So this is not Hungary regulating AI. It is a sectoral economic regulator asserting inspection rights over a learning model because that model determines a cost the regulator has to accept or reject. If you are tracking where algorithmic oversight actually attaches, it is worth noticing that this one attaches through cost recognition rather than through a rights framework.
What happens if the operator will not co-operate
Point 8.1.2 carries no express sanction. There is no fine in it, and we are not going to quote one that does not exist.
The decree's consequence mechanism is elsewhere and it is structural: the methodology permits non-recognition of costs in defined circumstances, turning on substantiation, methodology and reasonableness. Fines paid by the operator are themselves treated as non-recognisable costs. What the decree does NOT say is that refusal or inability to facilitate an 8.1.2 analysis disallows the reserve-procurement cost. We are not asserting that consequence.
So the honest statement is narrower than it first appears: the analysis right exists, the non-recognition machinery exists, and the decree does not join them with an express consequence for non-cooperation. Anyone who tells you otherwise is reading a clause that is not there.
Two things are worth carrying from this. First, an algorithmic audit right can arrive without any AI vocabulary at all: 8/2026 MEKH gives a regulator a standing right to analyse a learning algorithm on historical data while never once saying artificial intelligence or machine learning. If you are monitoring AI governance by keyword, you will miss instruments like this. Second, there is no fine anywhere near this provision. The decree's lever is cost recognition, but note carefully that it does not expressly tie non-recognition to refusing an 8.1.2 analysis, so do not plan around a consequence the text does not state. The decree is dated 28 August 2026 and in force from 30 August 2026.
Source File
Open Magyar Kozlony 2026. evi 120. szam and find 8/2026. (VIII. 28.) MEKH rendelet. Confirm the title refers to the modszertan for the eszkoz- es koltsegfelulvizsgalat preceding the arszabalyozasi ciklus starting prices, and that it is signed Dr. Juhasz Edit, elnok. Read Section 8 and confirm Ez a rendelet 2026. augusztus 30-an lep hatalyba. Then find point 8.1.2 and confirm the phrase oras bontasu tanulo algoritmus and the words historikus adatok. Search the whole decree for mesterseges intelligencia and gepi tanulas; both should return zero. One retrieval note: the gazette's /megtekintes path returns HTTP 200 with a small text/html viewer shell rather than the PDF. The file itself is behind /letoltes.
A Hivatal jogosult historikus adatok alapján részletes elemzéseket végezni a tartalékbeszerzési igények alapját képező, előrejelzések (PV-termelés, időjárás, rendszerterhelés) alapján pontos igénymeghatározást végző, órás bontású tanuló algoritmus működésével kapcsolatosan. - 8/2026. (VIII. 28.) MEKH rendelet, point 8.1.2
FAQ
Is this an AI regulation?
Not by its own terms. The decree never uses mesterseges intelligencia or gepi tanulas. It refers to a tanulo algoritmus, a learning algorithm, three times in a single provision. It is energy cost-regulation methodology that happens to assert an inspection right over a learning model.
What exactly can the regulator do?
Under point 8.1.2 the Authority is entitled to carry out detailed analyses, on the basis of historical data, of the operation of the hourly learning algorithm that determines reserve procurement needs from forecasts of PV production, weather and system load.
Who has to comply?
The transmission system operator, az atviteli rendszeriranyito, identified in the annexes as MAVIR. The right belongs to MEKH.
What is the penalty for non-compliance?
There is none attached to point 8.1.2. The decree operates through cost recognition, permitting non-recognition in defined circumstances that turn on substantiation, methodology and reasonableness. It does not expressly provide that refusing or failing to facilitate an 8.1.2 analysis disallows the reserve-procurement cost.
When did it take effect?
Section 8 provides that the decree enters into force on 30 August 2026. Its own date is 28 August 2026.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.