InfrastructureIssue 0110 records
Utility Management
First utility-management edition. State regulators are moving to inventory utility AI: Connecticut's AG and Consumer Counsel petitioned PURA for a full AI probe, and Arizona's commission held an AI guardrails workshop, echoing Ofwat's UK plan. Readiness evidence from NRECA, Ofwat and AWWA (49% of water respondents lack an AI policy) points to data quality and basic policy as the binding constraints. Deployment evidence is promising but unproven: New Mexico's satellite-AI program found 216 leaks with projected, not measured, savings; Genesee County reports faster condition assessments; Austin signed an AI water MOU with K-water. A UIUC preprint shows AI weather forecasts help pre-storm crew staging only up to about three days. The FBI/EPA PLC attack alert sets the security baseline before AI touches OT.
- Evidence records
- 10
- Cross-source patterns
- 3
- Evidence classes
- 3 independent reporting2 government evaluation2 standards or public-body guidance2 public-sector association guidance1 academic research
- Outcomes
- 6 emerging3 cautionary1 mixed
- Source freshness
- 4 older, newly relevant3 recent2 new this fortnight1 undated
- Research completed
- 2026-10-30
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Regulators are moving to inventory utility AI before setting rules
Connecticut advocates petitioned PURA to require full AI system disclosure with oversight levels; Arizona's commission opened a generic AI docket with a guardrails workshop; Ofwat began structured evidence gathering ahead of guidance. Following New York's archived September order, the common first step is an inventory, not a mandate.
Operating questionCould the utility produce a complete AI register, including vendor-embedded AI and AMI analytics, with owner, purpose and human-oversight level, if a regulator asked next quarter?
Supporting evidenceConnecticut Office of the Attorney General and Office of Consumer CounselArizona Corporation CommissionOfwat (Water Services Regulation Authority, England and Wales)
Data readiness and basic AI policy are the binding constraints
Ofwat's companies cited data quality as the top AI concern; NRECA lists identifier, timestamp, system-context and record-drift conditions; AWWA finds 49% of respondents lack an AI policy; Genesee County illustrates staff using general-purpose AI with SCADA and CMMS files.
Operating questionWhich single decision would AI support first, and do the identifiers, records and acceptable-use policy behind it meet a minimum readiness bar?
Supporting evidenceNational Rural Electric Cooperative Association (Business and Technology Strategies)Ofwat (Water Services Regulation Authority, England and Wales)American Water Works AssociationGenesee County Drain Commissioner's Division of Water and Waste Services (GCDC-WWS)
Detections and forecast skill are not the same as outcomes
New Mexico's 216 satellite-AI leak detections translate into projected, not measured, savings, and repairs are unfunded by the program; the UIUC study finds a 17.5% forecast-skill gain produced no significant crew-staging decision gain. Value claims must be tied to verified repairs or decision metrics.
Operating questionWhat closed-loop measure (verified repair, captured outage burden) will show the AI changed an outcome, and who owns collecting it?
Supporting evidenceNew Mexico Environment Department (with ASTERRA)University of Illinois Urbana-Champaign (Golkar, Alves de Oliveira, Alipour)
Evidence ledger
Full record · every source keeps its link and limitations
Forecast Lead Time and the Timing of Crew Deployment for Weather-Driven Power Outages Using AI Weather Models
Using GraphCast and FourCastNet operational forecasts with ORNL EAGLE-I outage records (2022-2024, 2,491 counties), the authors find outage-occurrence skill (AUC-PR) falls from 0.1207 at 24h to 0.0646 at 120h, still above a 0.047 county climatology. Staging crews in the top 10% of counties captured 0.356 of high-burden outage at 24h vs 0.238 at 120h; the 72h-to-120h loss is significant. Committing at 72h with all crews beat 24h with 60% of crews; break-even availability 83.4%. A 17.5% skill gain from land-surface predictors produced no significant decision gain, joining antecedent predictors at valid time inflated skill 42-53% (data leakage), and 90% prediction intervals covered only 0.476 of tropical cyclone events.
Why it matters, evidence and limitations
- Why it matters
- Directly relevant to municipal utilities, public power agencies and co-ops that request mutual-aid crews before storms and are being sold AI outage-prediction tools; it gives a decision-value test rather than an accuracy claim.
- Evidence and measured results
- Retrospective modelling study: LightGBM hurdle model, temporal split (2022 train, 2023 calibration/selection, 2024 test), block bootstrap over 18 storm weeks, paired lead-time comparisons, SHAP attribution, conformalized quantile regression. Baselines are county climatology and a perfect-foresight oracle (0.893). Code and data on Zenodo. Not peer reviewed.
- Limitations and uncertainty
- Preprint; three years of data with a single training year; 2024 test year had an unusual hurricane season; western winter skill unexplained; county/6h resolution unsuited to feeder-level restoration; monetary values highly assumption-dependent (VoLL, restoration acceleration).
Attorney General Tong, Consumer Counsel Coleman Petition PURA for Probe of Artificial Intelligence Use by Public Utilities
Connecticut's Attorney General and Consumer Counsel petitioned PURA to open a comprehensive review of utility AI use, asking that utilities report all AI systems, their objectives, level of human oversight and governing policies, and that PURA review data-security and cybersecurity protections and breaches. Requested scope covers storm response, customer systems, asset management, load and water-demand forecasting, procurement, grid/plant/water treatment operations, third-party vendors and AMI. The petitioners cite risks including hallucination and bias, and reports that conventional weather models outperformed AI models for a February nor'easter.
Why it matters, evidence and limitations
- Why it matters
- Signals that state regulators and consumer advocates may require AI inventories and human-oversight disclosures from electric and water utilities, following New York's September 2026 order already in the SLED archive.
- Evidence and measured results
- Advocacy petition and press release; it states concerns and requested scope but presents no new measured findings. Whether PURA opens the docket was not established at retrieval.
- Limitations and uncertainty
- Petition only; PURA action and timing unknown. Claims about AI forecast performance are cited by the petitioners, not independently evidenced in the release. Named utilities are investor-owned (Avangrid, Eversource), though water utilities are in requested scope. Classed government-evaluation as the closest schema class; it contains no evaluative findings.
Arizona Corporation Commission special open meeting: Researching and Discussing the Use of AI to More Efficiently and Reliably Deliver Energy and Water (Docket AU-00000A-26-0060)
The ACC held a 21 October 2026 workshop under a generic AI docket covering energy and water utilities, with an EPRI overview and best-practice presentations from Trico Electric Cooperative, Arizona Public Service, Nexus Water Group, Salt River Project and ACC Utilities Staff. The stated purpose was education on AI applications, best practices and guardrails; no substantive votes were taken.
Why it matters, evidence and limitations
- Why it matters
- Shows another state commission (alongside New York and the Connecticut petition) opening an AI-specific proceeding, notably including a co-op, a public power entity (SRP) and a water utility group.
- Evidence and measured results
- Agenda and notice only; presentations and any staff recommendations were not inspected. No findings or measured outcomes.
- Limitations and uncertainty
- Notice only; content of presentations unknown; publication date of the notice not stated; workshop is educational with no votes. Classed government-evaluation as the closest schema class.
New Mexico Looks to Space for Help Combatting Water Waste (LeakTracer year-one results)
NMED's statewide LeakTracer program uses ASTERRA L-band satellite radar plus AI to flag likely treated-water leaks, which are matched to utility GIS and verified by field inspectors. In year one, 28 utilities, 181 miles of pipe inspected and 216 leaks identified (45 main, 54 service, 117 customer-owned). Repairing all would recover an estimated 898,000 gallons/day (328 million gallons/year) and about $505,000/year in production costs, but NMED stresses these are projections, not measured savings; utilities pay for repairs and NMED is building a repair follow-up process.
Why it matters, evidence and limitations
- Why it matters
- A state-funded shared AI service for systems serving under 20,000 people, a model state agencies and small utilities can replicate; it also illustrates the gap between AI detections and verified outcomes.
- Evidence and measured results
- Reporting quoting NMED's communications deputy director and the state's Year One Summary; savings figures are estimates assuming all leaks are repaired. No independent verification of detection accuracy or completed repairs.
- Limitations and uncertainty
- Vendor-supported program; detection precision, false positives and repair completion rates not reported; savings are projections; trade press source.
With AI, Water Utilities Can Manage Assets More Closely
GCDC-WWS uses general-purpose AI (ChatGPT) to analyze SCADA and CMMS data trends and prepare grant applications (one reportedly cut from about a week to an afternoon), and uses MentorAPM on GIS for photo-based condition assessments, which the assistant director says could cut a system-wide assessment from months to days, with automatic work orders. It is also working with Fontus Blue on AI water-quality optimization. A CDM Smith/Trinnex leader cites workforce retirements, legacy systems and data quality as key challenges.
Why it matters, evidence and limitations
- Why it matters
- Concrete municipal water example of AI for asset management, condition assessment and grant writing, plus the workforce-retirement driver common to SLED utilities.
- Evidence and measured results
- Self-reported practitioner statements and a consultant's views; no measured before/after data or independent evaluation. Vendor names cited.
- Limitations and uncertainty
- Anecdotal time savings; vendor-mentioned products; no outcome measurement; the condition-assessment timing was clarified in an editor's note.
Malicious Cyber Actors Targeting Water and Wastewater Sector Internet-Facing Programmable Logic Controllers, Causing Operational Disruptions
Since 27 July 2026, water and wastewater utilities in at least seven states reported attacks on internet-exposed Rockwell/Allen-Bradley MicroLogix 1100 and 1400 PLCs; actors changed IP addresses and passwords, causing loss of view and in some cases control, and at least one victim found modified ladder logic. Reported effects included loss of pressure and flooding. Similar third-party network setups across victims may let attackers multiply successes. FBI and EPA recommend removing PLCs from internet exposure via secure gateways, strong unique passwords, access control lists, and securing cellular modems.
Why it matters, evidence and limitations
- Why it matters
- Any AI analytics or optimization that reads from or writes to utility OT expands the same attack surface; this alert sets the baseline controls municipal water systems must have before connecting AI to SCADA/PLC data.
- Evidence and measured results
- Federal PSA based on incident reports to the FBI; specific victims and counts per state not disclosed. The companion CISA alert reported boil water notices and sustained manual operations.
- Limitations and uncertainty
- The alert does not involve AI; it is included as a security precondition for AI-in-OT work. Dated July 2026, outside the immediate news window, but the campaign is recent and the first edition for this stream has no prior security baseline.
K-water signs MOU with Austin to share AI water management technology
K-water signed an MOU with the City of Austin, with a delegation led by Assistant City Manager Michael Rogers, to cooperate on AI water management: sharing technologies and best practices, joint seminars, and joint proof-of-concept projects for local application. Austin Water is interested in alternative water sources and smart management as urban development and data center expansion raise demand. K-water cites its digital twin platform and AI-based autonomous treatment plant operation system, the latter on track to become an ISO standard after a Draft International Standard vote.
Why it matters, evidence and limitations
- Why it matters
- A large U.S. municipal utility entering a government-to-government AI cooperation framework tied to data-center-driven demand growth, with proof-of-concept projects as the next step.
- Evidence and measured results
- News report largely reflecting K-water's announcement; no Austin-side statement, budget, scope or timeline; MOU is non-binding cooperation.
- Limitations and uncertainty
- Single source with K-water framing; MOU commitments are non-binding; no measured outcomes; commercial motive explicit (opening U.S. opportunities for Korean firms).
Data Science at Rural Electric Cooperatives, Part 1: Opportunities, Foundational Requirements, and Barriers
NRECA's first report in a four-part series frames data science (including ML/AI) for co-ops around decision value and readiness. It lists value areas (asset management, reliability and outage management, distribution planning, member programs/DER, financial decision support) and five foundational data conditions: consistent identifiers, temporal alignment, documented system context such as feeder and phase, records that reflect field reality, and sufficient history. It names staffing, recruitment and organizational adoption as non-technical barriers.
Why it matters, evidence and limitations
- Why it matters
- Member-owned co-ops are a core public-purpose utility segment; this gives a sector-authored readiness checklist that applies equally to municipal utilities.
- Evidence and measured results
- Guidance report by an NRECA data scientist; no case data or measured outcomes. Publication date is month-only (June 2026); 1 June used as the calendar date.
- Limitations and uncertainty
- Month-only publication date; Part 1 of 4 is conceptual; older than the current news window but included as the foundational co-op readiness reference for this first edition.
Infrastructure replacement, financing remain top challenges for water industry (2026 State of the Water Industry)
AWWA's 2026 State of the Water Industry survey (fielded late 2025) found 56% of respondents anticipate some positive impact from generative AI, while 49% say they have no policy governing AI use. Financial pressure frames adoption: only 43% are very to fully able to cover operating costs through rates and fees, 46% plan rate increases within five years, and 51% report high to considerable tariff impact on costs.
Why it matters, evidence and limitations
- Why it matters
- Quantifies the water sector's AI governance gap: optimism outpaces policy, under tight affordability constraints that shape what utilities can buy.
- Evidence and measured results
- Association survey summary; sample size and respondent mix not stated in the article; full report not inspected.
- Limitations and uncertainty
- Survey fielded in late 2025 and published April 2026; summary-level figures only; respondents may include non-utility members.
Ofwat AI Adoption Plan for the water sector during transition
Ofwat's informal evidence gathering with water company chief data officers and AI leaders found all reported live or embedded AI use. The top use cases were leakage and network operations, asset maintenance, billing/debt/customer support, contact handling and regulatory reporting. Wastewater uses include predicting sewer blockages and storm overflows from telemetry and rainfall. Poor data quality and AI data readiness was the most cited concern, workforce perceptions were flagged, and few companies were ready for regulatory sandboxes. Ofwat plans to develop guidance with the DWI and other regulators, deepen evidence gathering, and support innovation.
Why it matters, evidence and limitations
- Why it matters
- A peer regulator's structured approach to utility AI (inventory, guidance, sandboxes when ready) offers a template for U.S. commissions now opening AI inquiries.
- Evidence and measured results
- Regulator plan based on an informal, high-level engagement; no counts of respondents or quantified results.
- Limitations and uncertainty
- UK context with privatized water companies; informal evidence base; month-only date inferred from upload path; regulatory transition underway.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Government evaluation
- A public body’s measured evaluation or documented pilot.
- Standards or public-body guidance
- Normative or advisory guidance from a standards body or public institution.
- Public-sector association guidance
- Practitioner guidance or an association-supplied case; not independent outcome evidence.
- Academic research
- Research produced through an academic institution or peer-reviewed venue.