Insights

CNOOC Limited says its intelligent injection-production interaction scenario for offshore oilfield production was selected as a high-value scenario at the 2026 World Artificial Intelligence Conference. Under Kuwait Oil Company's Ahmadi Innovation Valley programme, Baker Hughes and SLB describe AI work touching water production, recovery from existing wells and water management, and KOC's own chief executive is quoted in one of the two releases, about technology deployment and knowledge transfer rather than about a field. Sinopec reports breakthroughs in synergistic oil flooding theories, and does not call them AI. Every one of those statements is about a capability. None of them is about the data that would make the capability identifiable, and that is a separate question with an arithmetic answer. A waterflood run to a voidage target moves its injectors together, and when injection schedules are correlated the posterior on a connectivity vector is narrow along the total support a producer receives and wide along the question of which injector supplies it. On a synthetic five-injector toy at a schedule correlation of 0.90 and 8% rate noise, one allocation fraction is known to 0.096 against a prior of 0.200 while the total is known to 0.042, an ambiguity of 5.76 times. Log bottomhole pressure at the producers and both widths improve while the ambiguity rises to 6.20, climbing toward the 6.78 the schedule alone allows. Only a designed perturbation moves that number, and the instrument prints how many weeks of it the tolerance costs.

An H-kappa stack can be re-read as a likelihood and reported as a calibrated credible region at essentially no extra compute. The reason to bother is not tidiness. As a station's data quality falls the region inflates along the trade-off ridge while the stack maximum barely moves, so a pipeline that reports picks shows almost nothing at all for a network that has lost its constraint.

In one fortnight Sinopec announced what it calls the industry's first digital expert, the Fenghuo industrial AI agent, alongside its Great Wall large model; PetroChina said it had vigorously implemented an Artificial Intelligence Plus initiative; and YPF said its renewed Corva agreement now aims at digital copilots and agents for well construction. Three operators, three announcements, and not one word about what the model reads. On the scans we digitised, the answer splits in two: the model resolves the header taxonomy and the depth scale reliably, and it recovered only 0.37 and 0.32 of the two curve masks against 0.97 on the background. So a digital expert answers a question about the label far better than a question about a value, and this piece computes by how much. The instrument prints the field count at which the two kinds of question are equally reliable, 38.4 on our opening settings, and that number does not move when you drag the stain or the skew: on an unrestored sheet both answers carry the same damage exponent, so the ordering is settled on the clean sheet. Deskew and stain normalisation are the only controls that move it, because they are the only ones that treat printed header glyphs and faint curve ink differently.

In four weeks of August 2026, ADNOC and SLB said an AI platform developed in the UAE and hosted within ADNOC's cloud environment had gone live across more than 120 drilling rigs, Baker Hughes said it would build a research and technology development centre for Kuwait Oil Company inside the Ahmadi Innovation Valley, Aramco and Aramco Digital said they had signed agreements on localizing critical technologies and building national industrial capabilities, and Sinopec said its Great Wall large model had further improved and its Fenghuo industrial AI agent had launched. Every one of those is a statement about where models will live. None of them carries the two numbers that decide whether the models can be kept current there: how many are in production, and how often each one is retrained. Put a fixed bench behind those two numbers and the arithmetic is unforgiving in a way that is easy to miss. On the on-prem bench we specified for one operator, at an assumed one retrain slot and an assumed 12 hour retrain, the fleet can grow 436% between the point where 5% of retrains miss the overnight turnaround from an illustrative scenario we published and the point where half of them do. Take the same bench to eight slots and that warning band closes to 23%. A larger sovereign bench is safer at any given fleet and tells you far less before it stops clearing the window.

Three national oil companies put AI on the record in August 2026 and published one set of performance figures between them, none of it production and none of it with a baseline: ADNOC with SLB on more than 120 drilling rigs, Baker Hughes speaking for Kuwait Oil Company at Ahmadi Innovation Valley, and CNOOC Limited on the Haineng-Zhiqing platform. The only production uplift numbers in the window belong to two small operators. U.S. Energy Development Corporation reports 12.2% against a 90-day pre-activation baseline on 12 Permian wells; Presidio Production Company credits about 1% to its Asset Intelligence Group. A pre-activation average is a counterfactual that no longer exists, because the wells would have declined anyway. Put the decline back into the arithmetic and the same measured 12.2% implies 12.2% on flat wells, 22.5% at 30% annual decline and 51.0% at 70%, all on the 90-day window the operator states. The instrument below computes both that surface and the smallest uplift the design could resolve, and shows the one lever that moves them apart at once.

A network trained on synthetic waveforms can score near-perfectly on a synthetic test set and then read a real station as noise. The failure is not a capacity shortfall, it is a distribution mismatch, and the fix is to measure the field's own noise floor, timing spread and gain variation and reproduce them channel by channel. Realism has an optimum rather than a direction, so overshooting costs as much as undershooting, and the only instrument that can tell you which side you are on is a held-out set of real labelled records.