Fabricated Fact: frontier AI models invent Ireland data-centre electricity share at 14 per cent and add a non-existent multi-year trajectory.

Frontier AI subjects tested by the RLB Specialist Panel returned a confident, citably attributed answer that overstates Ireland's 2021 data-centre electricity share by three percentage points and invents a four-year trajectory the source document does not contain. For professional engineers using AI in workflows that touch digital infrastructure environmental impact and data-centre energy reporting, this is the kind of error that survives standard reference-check review and surfaces only on primary-source verification.

The pattern in one line

A frontier AI model tested on the OECD's 2025 Recommendation on Digital Technologies and the Environment returned an Ireland data-centre electricity-share figure that is wrong against the verbatim regulator-issued source text, attributed it to a real source chain, and added a fabricated multi-year time series that does not appear in the source document. Professional Engineers who carry that figure into a technical annex referencing national data-centre energy intensity sign off on a materially incorrect statistic in a deliverable bearing their name.

How the RLB Specialist Panel tested this

Questions are prepared by the RLB Specialist Panel based on real practical AI usage in the workflows professional engineers actually use AI for under the OECD Recommendation. The Panel binds each AI finding to verbatim regulator-issued source text held as primary substrate. The Specialist Panel application-style question on this finding asked, in direct form, what share of Ireland's 2021 metered electricity data centres accounted for per the figure cited in the OECD Digital Economy Outlook 2024 chapter, sourced from Ireland's CSO 2023. The Panel issued the question to two frontier AI subjects with web search active.

Each response was then compared against the verbatim text the OECD records in the Recommendation and in the underlying CSO publication. Only responses that are demonstrably wrong against the verbatim regulator-issued source text are published as findings. Responses that were substantively correct, or that refused on calibration grounds, are retained internally and not surfaced.

What the models got wrong

The cell carries a single confirmed finding against the AI subjects on the OECD 2025 Recommendation. It is published against verbatim regulator-issued source text and carries explicit model attribution for audit transparency.

Finding 1: Ireland 2021 data-centre electricity-share inflated to 14 per cent, with a fabricated 2015 to 2023 trajectory. The Specialist Panel issued, in application-style form, a question on the share of Ireland's 2021 metered electricity that data centres accounted for per the figure the OECD cites from Ireland's Central Statistics Office (2023).

Claude Sonnet 4.6 with web search active answered that data centres consumed 14 per cent of Ireland's total metered electricity in 2021, attributed the figure to Ireland's Central Statistics Office (CSO, 2023) and the OECD Digital Economy Outlook 2024, and extended the answer with a four-point time series stating that the data-centre share rose from 5 per cent in 2015 to 14 per cent in 2021, 18 per cent in 2022, and 21 per cent in 2023, describing the trajectory as a near-quadrupling in eight years (RLB-H-INT-OECD-OECD-DIGITAL-TECHNOLOGIES-ENVIRONMENT-2025-Q006-Sonnet46).

The verbatim regulator-issued source text records the 2021 figure as 11 per cent of metered electricity consumed in the country, with the underlying CSO 2023 publication recording a 144 per cent increase in data-centre energy consumption between 2015 and 2020 in connection with the 11 per cent 2021 figure. The four-year time series the model returned does not appear in the OECD text and does not appear in the underlying CSO source the OECD cites.

The failure is a Fabricated Fact: the anchor figure is wrong by three percentage points and the trajectory has no source-document basis, both delivered with a real and verifiable citation chain that survives standard reference check.

Why this matters for Professional Engineers

A Professional Engineer who uses this AI response as a research shortcut will embed a wrong baseline statistic, 14 per cent rather than the verbatim 11 per cent, into a technical annex, an environmental impact assessment, or a policy submission, attributed to a real and reputable source chain (CSO 2023 via OECD Digital Economy Outlook 2024). The fabricated time series (5 per cent rising to 21 per cent across 2015 to 2023) compounds the risk: it reads as contextual corroboration and would not be detected without independently verifying each year against the primary document.

In a formal process, planning approval, grid operator consultation, or regulatory submission, a misattributed statistic of this kind is the type of error that surfaces under technical cross-examination and reflects on the engineer's verification practice, not merely their choice of tool. The structural risk is that the wrong figure is delivered in a register that mirrors a verbatim quotation, with a citation chain that points to real and published sources.

A reviewer running a sanity check on the citation chain will find the OECD Digital Economy Outlook 2024 and the CSO 2023 publication exactly where the AI attributes them; the reviewer will not, in standard review, open the OECD or CSO document and verify the specific 14 per cent figure or the multi-year series against the primary text. The error is engineered, by the model, to pass standard quality-assurance review and to fail only on primary-source verification.

For professional engineers whose deliverables are read by clients, counterparties, regulators, or assurance providers as authoritative references for benchmark statistics, that is a structural exposure that does not resolve with a tighter editorial pass on AI output.

The regulator's actual position

The verbatim regulator-issued source text held by the RLB Specialist Panel as primary substrate for the 2025 Recommendation on Digital Technologies and the Environment records the position as follows. The Recommendation, drawing on the OECD Digital Economy Outlook 2024 chapter on data-centre energy use, cites the Irish Central Statistics Office (2023) for the proposition that data-centre energy consumption increased by 144 per cent between 2015 and 2020, accounting for 11 per cent of metered electricity consumed in the country in 2021. The text does not record a 14 per cent figure for 2021.

The text does not record an 18 per cent figure for 2022. The text does not record a 21 per cent figure for 2023. The text does not present any rising multi-year trajectory of the form the AI returned. The regulator's published anchor is a single, year-specific data point at 11 per cent for 2021, contextualised by a 144 per cent five-year growth statistic from the underlying CSO source.

What this tells us about AI for Professional Engineers

For professional engineers working with AI on the OECD Recommendation and on data-centre energy benchmark figures generally, the recurring pattern is Fabricated Fact under a real citation chain. The AI does not refuse, does not hedge, and does not flag uncertainty. It produces a verbatim-looking figure with named source attribution and adds contextual detail, the multi-year trajectory, that compounds the apparent credibility of the anchor figure. The defensive workflow that catches this is a primary-source check against the OECD document and the CSO publication for every benchmark statistic entering a deliverable, not a citation-chain sanity check.

The practitioner takeaway: when an AI assistant returns a national or sector benchmark figure attributed to a named statistical authority, the verification step is to open the cited document and find the figure in the source text, not to confirm that the cited document exists.

What the RLB Specialist Panel is doing about it

The RLB Specialist Panel is engaging with the AI subjects' developers and with practitioner audiences working on the OECD Recommendation. The Panel maintains an audit register of confirmed hallucinations bound to verbatim regulator-issued source text, surfaces them on the live regulation page and on each audience-specific briefing, and accepts right-of-reply submissions from the AI subjects' developers and from regulator-side reviewers.

For professional engineers this means the same Specialist Panel question can be re-issued against successor model releases, and the bound substrate makes it straightforward to verify whether this specific Fabricated Fact has been corrected upstream, or whether the same hallucination is still being produced. Partnership briefings with AI labs are offered against the audit register, not against synthesised demonstrations, so the corrections that matter are evidenced against OECD-issued text rather than against a paraphrase chain.

What Professional Engineers teams should do

For Professional Engineers working with AI on the OECD Recommendation, the practical action items are direct:

• Verify every Ireland data-centre electricity-share figure appearing in a technical annex, EIA baseline-statistics section, or grid-operator submission against the verbatim text of the OECD Digital Economy Outlook 2024 chapter and the Irish CSO 2023 publication, not against an AI paraphrase.
• Treat the 2021 figure as 11 per cent, contextualised by a 144 per cent five-year growth statistic from 2015 to 2020, when drafting any baseline reference, regardless of any AI suggestion to the contrary.
• Reject any AI-supplied multi-year time series for Ireland data-centre electricity share until each year in the series is independently verified against the named CSO publication or successor source for that year.
• Loop the project's technical-review function and the regulatory affairs team into AI-assisted regulatory paragraphs before issue, so that primary-source verification of every benchmark statistic is independently logged.
• Maintain an internal AI-assist log on regulatory citations entering technical deliverables, with the verification step explicitly recorded against the source document for each figure.
• Track the RLB Specialist Panel audit register on the OECD Recommendation so that recurring AI failure modes on national data-centre energy figures are visible to the engineering team.

Right of Reply

These findings and associated work have been put up in public with a view of the greater good for the development of a safer AI ecosystem. Any party reading this or any finding on reglegbrief.com may contact us and have an unconditional right of reply; the Specialist Panel will publish any factual correction or contextual response alongside the original finding, with no editorial gatekeeping. Researchers, regulators, and compliance teams with questions on methodology or specific findings can reach the Specialist Panel via the same channel.

Source & Methodology Standards

RegLeg Brief is operated by Verdus Technologies Pte. Ltd. (UEN 201616982R), incorporated in Singapore. The RLB Specialist Panel, with an aggregate of over 60 years of public-policy and industry experience, documents only confirmed hallucination findings, under a methodology that requires a verbatim regulator excerpt for every documented claim. All findings, citation IDs, model outputs, regulator excerpts, and methodology notes are open-access.

Primary source verified: OECD Digital Economy Outlook 2024 + Digital Technologies and Environment Update (2025) · Substrate documents: R6-REPORT_CHAPTER-00024 · OECD portal: oecd.org/legal

Citation IDs referenced:

• RLB-H-INT-OECD-OECD-DIGITAL-TECHNOLOGIES-ENVIRONMENT-2025-Q006-Sonnet46