Is Gas-Supply AI High-Risk Under the EU AI Act? Annex III Point 2

It depends on what the model touches. AI used as a safety component in the supply of gas — the systems that hold pressure, regulate flow, and trip on a leak — is high-risk under Annex III point 2 of the EU AI Act, Regulation (EU) 2024/1689. AI that forecasts demand, optimises procurement, or runs your billing is generally minimal-risk. The same gas utility can own both, in two different tiers, and the difference is the phrase "safety component."

If you operate a gas distribution or transmission network, this page draws the line precisely — and tells you what the high-risk part owes, and when.

---

What Annex III point 2 actually says

Point 2 designates as high-risk AI intended to be used as a safety component in the management and operation of critical digital infrastructure, road traffic, or the supply of water, gas, heating and electricity. The supply of gas is named on the face of the provision. But the qualifier governs everything: the AI must be a safety component, meaning (under Article 3(14) as rewritten by the Digital Omnibus) it performs a safety function — preventing or mitigating risks to the health and safety of persons or property — or its failure or malfunction would endanger health and safety. AI used solely for efficiency, optimisation or convenience is not a safety component.

So the test is not "is this AI in a gas company?" It is "does this AI perform, or back-stop, a safety function in delivering gas?" Read the use, not the sector.

The line that decides it

Walk the gas value chain and the boundary becomes concrete. A control model that regulates line pressure, modulates flow at a regulator station, drives emergency shut-off, or interprets sensor data to detect a leak is a safety component — its malfunction can vent or ignite gas and endanger people. That is high-risk. By contrast, a model that forecasts seasonal demand for capacity planning, optimises wholesale procurement, schedules meter reads, or detects billing anomalies sits in the back office. Its failure costs money, not lives. That is minimal-risk, with no mandatory obligations under the Act.

A useful sanity check: ask whether the model's failure would, on its own, endanger people or property. If yes, treat it as a point 2 safety component until you can document otherwise. Demand and billing analytics fail that test; real-time safety control passes it.

What high-risk gas-supply AI owes

Where a system is a safety component under point 2, the full high-risk regime applies. The provider — whoever builds the control system and places it on the market under their own name (Article 16) — owes a risk-management system (Article 9), data governance (Article 10), technical documentation (Article 11, content set by Annex IV), logging (Article 12), human oversight (Article 14), and accuracy, robustness and cybersecurity (Article 15 — the requirement that bites hardest for safety-critical control). Conformity is assessed through the Annex VI internal-control route under Article 43 — a notified body is not generally required outside the biometrics category — followed by the Declaration of Conformity (Article 47) and CE marking (Article 48). Critical-infrastructure systems under Annex III point 2 are excepted from the public EU database and instead register at national level (Article 49(5)).

The gas operator is usually the deployer and carries the Article 26 duties: run the system per the provider's instructions, assign competent human oversight, monitor operation, and keep the automatically generated logs. Two parallel regimes matter here and are easy to miss: NIS2 cybersecurity obligations for essential entities, and the sectoral gas-safety rules you already comply with. The AI Act sits on top of them, it does not replace them.

The deadline, stated carefully. Stand-alone high-risk Annex III obligations were originally timetabled for 2 August 2026. The Digital Omnibus — Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026 and in force since 27 July 2026 — moved that to a fixed 2 December 2027. That is now the legal date. Plan against 2 December 2027.

Worked example: Stadtgas Rhein-Ahr GmbH

Stadtgas Rhein-Ahr GmbH is a German municipal gas distributor, ~210 employees, ~€44M turnover. It runs three AI tools, bought from two vendors.

1. A pressure-and-flow control model at its regulator stations that modulates output and triggers an over-pressure trip. This is a safety component under Annex III point 2 — high-risk. Stadtgas is the deployer; the vendor is the provider. Stadtgas must assign Article 14 human oversight, keep the logs, and confirm the vendor has done the Article 9–15 work and the Annex VI conformity assessment.
2. A demand-forecasting model that sizes winter procurement. Not a safety function — minimal-risk, no mandatory obligations.
3. A billing-anomaly detector that flags suspected meter tampering. Also back-office — minimal-risk.

Stadtgas scopes the high-risk stack around the control model alone, leaves the other two as the minimal-risk systems they are, and documents the split so a market-surveillance authority can see the reasoning. No Article 27 FRIA is triggered: Annex III point 2 systems are expressly excluded from it. As an SME, any fine would be capped at the lower of the percentage or the fixed figure under Article 99(6).

How Confir helps

Confir registers each gas-supply AI use, derives the role (deployer under Article 26, provider under Article 16 if you build the control system), and applies the safety-component test from Annex III point 2 to separate high-risk safety control from minimal-risk demand and billing analytics. For the systems that qualify it assembles the provider and deployer obligations, with the Article 15 robustness-and-cybersecurity emphasis safety-critical control demands. The engine is deterministic and rule-based — each finding is cited to the Article or Annex point that fired, and the same intake always produces the same result.

Related guides

• Critical infrastructure AI under Annex III point 2
• Electricity-grid AI and the safety-component test
• Water-supply AI under the EU AI Act
• How Article 6 classifies high-risk AI
• Accuracy, robustness and cybersecurity (Article 15)
• EU AI Act for energy and utilities