Black gold

Abu Dhabi’s ADNOC entrusts its oil operations to the algorithm developed by OpenAI and G42

The agreement between the UAE-based company and the two artificial intelligence firms is intended to support field planning, industrial operations and commercial decisions. It will guide decision-making even before an executive approves them. In the background are the European deadlines set out in the AI Act

by 24Ore NextMed

Aggiungi Il Sole 24 Ore
ai preferiti su Google
Pompe di estrazione del petrolio  REUTERS/Fabian Bimmer

4' min read

Translated by AI
Versione italiana

4' min read

Translated by AI
Versione italiana

On 2 November 2026 Adnoc, Abu Dhabi National Oil Company, announced a multi-year agreement with OpenAI and G42 to integrate artificial intelligence into the planning of oil fields, industrial operations and commercial decisions relating to the UAE’s oil sector. This is a platform that selects scenarios, compares investments and coordinates production, refining and sales, and guides decision-making even before a manager approves them. The company retains responsibility for the supply chain, but to exercise this responsibility it needs an IT infrastructure that it does not possess. From this point onwards, the issue is no longer merely one of productivity, but of sovereignty and economic security.

What does the agreement entail?

ENERGYai, developed within the ADNOC-AIQ ecosystem, utilises a 70-billion-parameter model and over fifty years of corporate expertise. In January 2025, ADNOC announced that the trial would utilise data from more than 15 per cent of its onshore and offshore wells . In its press release, the company highlights more than 200 AI applications and 70 robotic use cases, either operational or under development; over 40,000 employees have been provided with ChatGPT Enterprise, with a reported adoption rate of over 97 per cent. The Frontier platform adds the capability of agents connected to corporate systems, whilst G42 provides computing infrastructure in the UAE.

Loading...

In August 2026 Adnoc and SLB announced a platform for operations across more than 120 drilling rigs, claiming it would reduce the engineering workload by 30–40 per cent: a result relating to a specific system, not to Frontier. The aim of reducing the time taken to prepare development plans from months to hours also remains to be seen in terms of results: speeding up the analysis does not equate to reducing total costs, increasing revenue or improving safety.

Capacity and production: the barrel that remains underground

This distinction is crucial in the oil sector, where production capacity and actual output do not coincide. ADNOC reports a capacity of approximately 4.85 million barrels per day and a target of 5 million by 2027; neither figure constitutes a production order. Decisions must take into account the characteristics of the reservoir, the availability of facilities, investment requirements and coordination commitments between producers. For an algorithm, a barrel still underground is not simply equivalent to today’s price: its value depends on future costs, extraction times, transport, refining and alternative uses of capital. This is where the integration of AI can create value, but only if it makes explicit the assumptions underpinning each recommendation.

The 2026 market highlights just how fragile forecasting shortcuts can be. In its October report, the International Energy Agency estimated a decline in global demand of 2.6 million barrels per day for the year and a recovery of around 2 million for 2027; it reported global production of 102.3 million barrels per day in September and an average price for North Sea Dated of $116.89 per barrel in the same month. The shortage of refined products and transport disruptions are affecting actual revenue. Under these conditions, the quality of decisions made under uncertainty matters more than the ability to predict the price of crude oil.

Energy networks

The shift to energy networks alters the sequence of consequences. In refining, an incorrect analysis can jeopardise planning; in the electricity grid, a recommendation that fails to take account of reserves, stability and transmission constraints may become incompatible with operational safety. In the water and desalination sectors, a planning error can disrupt service continuity and quality; in the nuclear sector, as outlined in the International Atomic Energy Agency guidance published in September 2025, the adoption of AI requires a specific assessment of safety, data and human factors.

The same logic applies across sectors that are regulated separately. The telecommunications sector is moving towards objective-oriented network management, as set out in ITU-T Recommendation M.3043 of October 2025; finance and payments depend on authorisations, liquidity and shared services; defence logistics links maintenance, stockpiles and operational readiness. In shipping, the non-binding code on autonomous ships adopted by the IMO in May 2026 distinguishes between automation and command responsibility.

Systemic risk does not require a single entity to control a power station, a bank and a port: it is sufficient for different organisations to share digital identities, computing infrastructure or faulty data. Apparent redundancy may conceal a single dependency. This is why the European AI Act is more selective than the debate suggests: in the consolidated version of July 2026, it does not automatically classify all software used in critical infrastructure as high-risk, because what matters is the function and any security role it may play. The main requirements of Annex III take effect from 2 December 2027, whilst those relating to Annex I take effect from 2 August 2028. Article 14 requires, where applicable, effective human oversight; Articles 16 and 26 distinguish between the obligations of suppliers and users. However, a human signature on a decision that nobody can reconstruct does not restore control. The same applies to contracts between developers, integrators, computing service providers and industrial operators.

What will be decided between November 2026 and November 2028

Between November 2026 and November 2028, European deadlines for high-risk investments will coincide with pressure to demonstrate returns on investment in infrastructure and automation. The test will be the ability to revoke permits, review decisions, migrate processes and maintain energy, water, communications and financial services during a failure. Governments and businesses will have to decide which operators can make proposals, which can carry out reversible activities, and which should be excluded from critical control functions. Above all, they will have to address the progress already made in the Gulf.

Copyright reserved ©
Loading...

Brand connect

Loading...

Newsletter

Notizie e approfondimenti sugli avvenimenti politici, economici e finanziari.

Iscriviti