InfrastructureIssue 116 records
Utility Management
Security and operations both showed AI's real shape tonight. Dragos's Senate testimony, recirculated this week, describes an AI model that found a Monterrey water utility's OT gateway on its own and failed only after about 2,800 rejected logins, and a gas utility where an AI assistant knocked its own access offline. Indiana utilities describe AI speeding attacks while small systems struggle with new state evaluation duties. On operations, Goulburn-Murray Water's in-house AI pump monitoring let staff stop a failing pump within eight minutes; a K-water-backed leak pilot in Sarawak confirmed 23 leaks in four days against preset targets; Spain's Nedgia began a staged ML gas-pressure pilot; and a Quebec firm outlined 10–12 week AI flood-forecasting pilots. NARUC's water-cyber vote is still underway. U.S. billing AI, stormwater results and co-op news were thin.
- Evidence records
- 6
- Cross-source patterns
- 2
- Evidence classes
- 5 vendor claim1 independent reporting
- Outcomes
- 4 emerging1 cautionary1 mixed
- Source freshness
- 5 new this fortnight1 recent
- Research completed
- 2026-11-10
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
AI speeds attackers toward utility OT boundaries, but credentials, isolation and segmentation are still what stop them
Dragos's testimony describes an AI model that found a municipal water utility's OT gateway on its own, only to fail after about 2,800 rejected logins, and a gas utility where an AI assistant knocked its own access offline. Indiana utilities interviewed this week describe the same picture: AI is accelerating attacks and phishing, while OT isolation and layered controls remain the defence, and small systems lack staff. Together they suggest utilities should prioritise boundary credentials, OT isolation and monitoring before AI-specific tooling. This is interpretation across vendor testimony and local reporting, not measured outcomes.
Operating questionAre all IT-reachable OT gateways and site firewalls free of default or single-password access, and would a burst of failed logins be noticed within minutes?
Supporting evidenceDragos, Inc. (testimony by Kurt Gaudette to U.S. Senate HSGAC Subcommittee on Disaster Management, District of Columbia, and Census)Daily Journal (Johnson County, Indiana); sources include Indiana Department of Environmental Management, Citizens Energy Group, Purdue/IU Cybertrack, Ice Miller, Tipton Municipal Utilities
Water and gas network AI pilots this week keep people in the loop and stage from monitoring to action
G-MW's pump monitoring alerts staff, who decide whether to stop a pump. The Sri Aman leak pilot uses AI to narrow where crews survey, with every flagged point checked in the field against predefined targets. Nedgia's gas pilot starts with monitoring, then remote set-points, and only later an ML optimiser. The common design is AI as a prioritiser with staged authority and field verification. None of these sources provides independent evaluation, so this is an observed design pattern, not proof of effectiveness.
Operating questionFor each operational AI pilot, what decision does the AI make versus recommend, and how are false alarms and misses measured before authority expands?
Supporting evidenceGoulburn-Murray Water (G-MW), Victoria state-owned rural water authorityWI.Plat (vendor) for JBALB Sarawak under K-water Export-Led Pilot ProcurementNedgia (Naturgy group) with Utonomy; article sponsored by Naturgy in La Razón
Evidence ledger
Full record · every source keeps its link and limitations
Dragos Senate testimony: AI tools steered intrusions toward a municipal water utility's OT gateway and knocked a gas utility's firewall offline, but neither reached OT
Dragos testified that, in an early-2026 intrusion into the enterprise IT network of Servicios de Agua y Drenaje de Monterrey (SADM), the attacker's AI model (Claude) found an internally accessible OT interface without being directed to, judged it high-value and ran a tailored credential attack; about 2,800 login attempts were rejected and no industrial system was reached. In a separate case at an unnamed gas utility, an attacker using a commercial AI coding assistant with valid admin credentials on a firewall saw the assistant make a configuration change that took the device offline, disrupting the utility's site-to-site connectivity, with no evidence of ICS impact. Dragos says AI lowers attacker skill and cost barriers but has not raised the ceiling against well-defended OT, and it knows of no confirmed AI-generated malware causing physical impact.
Why it matters, evidence and limitations
- Why it matters
- Why now: this September testimony was recirculated in UK coverage on 5 Nov and sits alongside NARUC's water-cybersecurity resolution, which is up for a vote 8–11 Nov. For municipal water and gas utilities it is first-hand evidence that the IT/OT boundary, credentials and remote-access devices are where AI-assisted attackers end up, and that basic controls held.
- Evidence and measured results
- Written testimony from a vendor that analysed operator logs and artifacts; the Monterrey case is also described in Dragos's own blog (1 Sep 2026, 350+ artifacts). Quantitative detail: ~2,800 rejected logins; Dragos tracks at least 26 OT-capable threat groups. No independent verification of the gas-utility case; the utility is unnamed. Testimony date taken from the hearing materials; the PDF itself carries no clear date line.
- Limitations and uncertainty
- Vendor testimony with commercial interest; cases are a small number of investigations; the gas utility is unnamed; the source predates this run and is included because it resurfaced in current coverage and bears on this week's NARUC vote.
Indiana water utilities weigh AI-accelerated cyber threats as SEA 459 assessment and reporting duties fall hardest on small systems
Reporting on Indiana water and wastewater cybersecurity: IDEM says some utilities saw outside activity on their networks this summer with no operational disruption. Indiana's 2025 Senate Enrolled Act 459 requires covered utilities (serving at least 500 people, plus certain treatment works) to run annual cybersecurity evaluations, name a state contact and report incidents. Citizens Energy Group's security director says AI is accelerating both attacks and defence ('a bit of an AI arms race'), that the utility uses AI agents to comb data, and that it isolates OT from the internet with layered controls. Cybertrack (Purdue/IU) warns that corrupted monitoring data can drive incorrect responses. Ice Miller notes about 80% of water infrastructure is run by local governments with limited resources. Tipton Municipal Utilities, attacked in 2024, says SEA 459's impact was minimal because it already ran evaluations.
Why it matters, evidence and limitations
- Why it matters
- Shows a state mandate in operation and the practical gap between large utilities with security staff and small municipal systems. Relevant to state programmes, shared-service offers and to how utilities frame AI in security operations.
- Evidence and measured results
- News article dated 8 Nov 2026 with named sources; no measured outcomes or incident statistics beyond qualitative statements. The 80% local-government figure is attributed to an attorney, not a dataset.
- Limitations and uncertainty
- Single-state local reporting; no data on compliance rates or outcomes; AI claims are practitioner opinion.
Goulburn-Murray Water's in-house AI condition monitoring flagged critical pump vibration and staff stopped the pump within eight minutes
G-MW's managing director writes that the authority's Electrical and Mechanical Services team built a condition monitoring system by fitting vibration sensors to pump motors at Tungamah pump station, linking them to motor and power monitoring panels and to AI that flags unusual patterns for staff. During the pilot it detected critical vibration on Main Pump 1; staff assessed the alert and stopped the pump before catastrophic failure, all within eight minutes. He stresses staff remain responsible for decisions and that G-MW is also investigating drones, LiDAR, aerial imagery, AI and remote survey boats to prioritise channel maintenance.
Why it matters, evidence and limitations
- Why it matters
- A public water authority building AI condition monitoring in-house with staff-in-the-loop is a realistic pattern for U.S. municipal and irrigation pump stations, where unplanned pump failures drive cost and service risk.
- Evidence and measured results
- Operator-authored opinion piece (self-reported); one incident, no accuracy, false-alarm or cost data; no vendor named. Evidence class set to vendor-claim as the closest class for self-reported operator claims.
- Limitations and uncertainty
- Single anecdote from the authority's own leader; no evaluation method, baseline or false-positive information.
K-water export pilot in Sri Aman, Malaysia: AI-screened acoustic survey confirmed 23 leaks in four days, vendor reports targets met
WI.Plat reports a Sep–Oct 2026 pilot for JBALB Sarawak with 75 smart sensors and handheld acoustic devices. 4,004 leak-sound records were collected; AI flagged suspected leak areas for a focused second survey. In four days the team confirmed 23 leaks, with repairs at 16 points checked on site. Reported targets met: leak detection accuracy 95%+ on 2,063 held-out records, 97.96% data collection rate, and 100% field match for AI-suspected points.
Why it matters, evidence and limitations
- Why it matters
- A government-backed export pilot with stated targets and a held-out test set is a more structured template than typical vendor claims; useful for U.S. utilities designing AI leak-detection pilots with acceptance targets.
- Evidence and measured results
- Vendor case study dated 6 Nov 2026. Accuracy measured on 2,063 records (30% of 6,876 from 50 sensors); field-match metric counts confirmed AI-suspected points but does not report missed leaks. No water-loss volumes or costs reported.
- Limitations and uncertainty
- Vendor-authored; 100% field match says nothing about false negatives; Malaysian network conditions differ from U.S. systems.
Spain's largest gas distributor starts ML-assisted pressure-control pilot to cut leakage; sponsored piece cites UK and Stockholm results
Sponsored La Razón article (5 Nov 2026) says Nedgia has started a pilot in Igualada with Utonomy to adjust low-pressure network pressure to real-time demand using sensors, remote communications and AI, retrofitted on existing regulators. It cites prior results: an SGN/Utonomy six-month trial in southern England (Oct 2024–Mar 2025) cut regulator pressure by up to 31% with a projected 16% average methane reduction on equipped sections, and Gasnätet Stockholm cut average night pressure from 15 to 12 mbar, about 20% lower losses. Utonomy's release says the six-month pilot is two months of installation and data generation then four months of operation and validation, starting with monitoring and remote set-points before a machine-learning optimiser.
Why it matters, evidence and limitations
- Why it matters
- U.S. municipal gas utilities face methane and leak-reduction pressure. Dynamic pressure management with ML is a measurable, retrofit approach worth tracking; this fills a gas-distribution gap in recent coverage.
- Evidence and measured results
- Sponsored content and vendor release; prior-trial figures are cited, not independently verified here; Igualada pilot results not yet available.
- Limitations and uncertainty
- Sponsored by the operator's parent; figures from vendor trials; European regulation and network design differ from U.S. systems.
Quebec engineering firm pitches 72-hour AI urban flood forecasting for municipalities, recommending 5–6 sensor pilots
Avizo and CANN Forecast presented InteliFlow on 4 Nov 2026: a platform combining weather models, rain gauges and local level/flow sensors with AI to forecast water levels and flows up to 72 hours ahead. Sensor data is quality-controlled before modelling, alerts are configurable to municipal thresholds (24/48/72 h), a shared dashboard serves water, public-safety and communications teams, and data is owned by the municipality and exportable for PCSWMM calibration. Avizo recommends a targeted pilot of five to six sensors over 10 to 12 weeks.
Why it matters, evidence and limitations
- Why it matters
- Gives U.S. stormwater managers concrete pilot sizing and data-ownership terms for AI flood forecasting, an area thin in recent coverage. No measured results are reported.
- Evidence and measured results
- Vendor webinar recap; no accuracy, lead-time verification or deployment outcomes.
- Limitations and uncertainty
- Marketing material; Canadian context; no performance data.
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.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.