Industries / Retail, hospitality and personal services / Retail and e-commerce / Convenience stores and fuel retail
Underground tanks and leak checks
This work keeps the fuel tanks buried under a station from leaking, finds leaks early when they start, and follows the rules that govern both.
7
Milestones achieved, from 1984 to 2026.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1984
Congress creates the tank program
In November 1984 Congress added Subtitle I to the Solid Waste Disposal Act, through the Hazardous and Solid Waste Amendments, and created the first national program for underground storage tanks (USTs), after a 1983 CBS 60 Minutes report on families in Canob Park, Rhode Island, hit by leaking gasoline; in 1985 the Environmental Protection Agency (EPA) estimated there were about 2.1 million such tanks, and in 1986 Congress added a trust fund paid by a 0.1 cent tax on each gallon of motor fuel. By March 2026 EPA counted 534,189 active regulated tanks at about 190,224 facilities, and 583,313 releases confirmed by states since 1984.
Credit
US Congress; EPA Office of Underground Storage Tanks (formed September 1985)
1988
EPA sets leak rules
EPA's September 1988 technical rule told owners to prevent leaks, detect them quickly and clean them up, required leak detection on all older tanks by 1993, and gave existing tanks until 22 December 1998 to add spill, overfill and corrosion protection or be closed.
LessonA hard deadline reshaped the tank population, because owners who could not upgrade had to close, and the Association of State and Territorial Solid Waste Management Officials (ASTSWMO) says the 1998 deadline "had a significant impact on the composition of the nation's UST infrastructure.".
Credit
EPA Office of Underground Storage Tanks
2005
Inspections every three years
The Energy Policy Act of 2005 amended the tank law to require an on-site inspection of every regulated tank at least once every three years, and added operator training, delivery prohibition (stopping fuel deliveries to tanks that break the rules), and secondary containment (a second wall that catches a leak from the first). It also required states to report publicly the source and cause of releases.
Credit
US Congress; EPA
2011
Japan's rule for old tanks
From 1 February 2011 a ministry ordinance (Ministry of Internal Affairs and Communications Ordinance No. 71 of 2010) sorted single-walled steel tanks buried directly in the ground by age, outer coating and wall thickness, and required those at "especially high" risk of corrosion to get an internal fiberglass lining of at least 2 mm or cathodic protection (a small electric current that stops rust), with a deadline of 31 January 2013. The ministry's own assessment said spills from hazardous-materials sites had reached 386 in 2008, that corrosion and wear caused about 40% of them, and that half of those came from underground tanks and pipes.
LessonThe cost fell on small stations; the private research firm Teikoku Databank linked the 2013 deadline to station closures, counting 252 fuel retailers that failed, closed or dissolved in the 2012 financial year, and Japan's energy agency now counts 381 municipalities with three or fewer stations (end of financial year 2024), citing aging facilities among the causes.
Credit
Fire and Disaster Management Agency (FDMA)
2015
EPA tightens upkeep checks
EPA's revised rule, published 15 July 2015 and the first large change since 1988, added walkthrough inspections every 30 days, tests of spill buckets and containment sumps at least every three years, inspection of overfill devices every three years, secondary containment for new and replaced tanks and pipes, operator training and checks that tanks can safely hold biofuel blends, with the main deadlines on 13 October 2018.
LessonRules on paper are not the same as practice; for April 2025 to March 2026 EPA reports that only 60.9% of inspected facilities met every technical requirement.
Credit
EPA Office of Underground Storage Tanks
2015
Tank gauges found online
On 10 January 2015 Rapid7 scanned the whole internet and found approximately 5,800 automatic tank gauges (ATGs, electronic probes and consoles that measure the fuel in each tank and run leak tests) reachable with no password, more than 5,300 of them in the United States, and warned that an attacker could shut down a station.
LessonA device installed to catch leaks is also a network device, and it needs the same protection as any other.
Credit
HD Moore, Rapid7; Jack Chadowitz of Kachoolie, who asked Rapid7 to look
2026
Attacks on gauges confirmed
On 2 June 2026 eight US agencies, among them the Cybersecurity and Infrastructure Security Agency (CISA), the Federal Bureau of Investigation (FBI) and EPA, said attackers were breaking into internet-exposed tank gauges and changing them by running commands, which can alter tank settings and switch off the alarms that warn of leaks; they told owners to take the gauges off the internet and set strong passwords. In 2024 Bitsight had disclosed 11 vulnerabilities in six gauge models from five makers, coordinated with CISA, and had counted 6,542 gauges online with no security code.
LessonA device that guards against leaks can be switched off from outside when it sits on the open internet, and research warnings in 2015 and 2024 did not by themselves get the gauges taken offline.
Credit
CISA and partner agencies; Pedro Umbelino, Bitsight
Watch for
Fewer exposed gauges online
Bitsight says the number of US internet addresses showing an exposed tank gauge, which sat between about 4,300 and 5,300 for ten months, fell to 3,850 in April, 2,624 in May and 2,354 in June 2026, 56% below the March peak, in step with industry alerts from mid-April; it says some devices may only have moved behind firewalls or virtual private networks.
Achieved whena second, independent internet survey with a published method confirms the fall, and the lower count holds for at least twelve months.
Claimed by
Pedro Umbelino, Bitsight, August 2026
Open
01
Which tank features predict leaks
A public, repeatable analysis that joins state tank records (install date, tank and pipe material, single or double wall, leak-detection method, corrosion protection) to the same states' confirmed-release records, and estimates how much each feature changes the chance of a release, with all code and data published.
The briefCredit · unclaimed02
Code the cause of releases
A labeled public set of state release records, and an AI classifier, that records for each release where it came from (tank, pipe, dispenser, spill bucket, delivery, other) and why (corrosion, physical damage, poor installation, spill, overfill, other), so that state counts can be compared.
The briefCredit · unclaimed
Open 01 · the brief
Which tank features predict leaks
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, repeatable analysis that joins state tank records (install date, tank and pipe material, single or double wall, leak-detection method, corrosion protection) to the same states' confirmed-release records, and estimates how much each feature changes the chance of a release, with all code and data published.
Why it is still open
ASTSWMO's Aging Tanks Workgroup found in 2015 that "there is no uniform set of criteria for evaluating risk of release" and that most states do not even calculate the average age of their tanks, although all 27 responding states record tank material and wall type and 26 record the leak-detection method and whether a release occurred. EPA's UST Finder (version 1.3, December 2025) lets anyone download facility and release locations, but EPA describes no analysis linking tank features to releases.
Built with what exists
UST Finder downloads; state tank and release files from state open-data portals; Python with pandas, scikit-learn and the lifelines library for survival analysis (time until a first release); an AI coding agent whose work is checked by tests (see the Note "Code no person read").
First step
Download the UST Finder facility and release layers and one state's tank-level file, and count how many facilities can be matched between them by facility ID.
Done looks like
For at least three states, a survival model of time to first release for each tank system, with published effect sizes and 95% confidence intervals for age, material, wall type and leak-detection method, checked on a fourth state held out from fitting, and rebuilt from the raw downloads by one command.
Open 02 · the brief
Code the cause of releases
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A labeled public set of state release records, and an AI classifier, that records for each release where it came from (tank, pipe, dispenser, spill bucket, delivery, other) and why (corrosion, physical damage, poor installation, spill, overfill, other), so that state counts can be compared.
Why it is still open
ASTSWMO reports that most states record source and cause but many call their data "inadequate and/or incomplete," that "No comprehensive source of this vital information is known to exist," and that its own study did not analyze the source-and-cause data states must publish under the 2005 Energy Policy Act.
Built with what exists
Gemini Decision 1, which returns probabilities for many fixed-answer questions per record (see the Note "A thousand questions, one pass"); pdftotext or OCR for scanned reports; a spreadsheet for hand labels.
First step
Pick two states that publish release reports with a written description, download 200 records, and label source and cause by hand.
Done looks like
1,000 public release records from at least two states, each labeled by two people with their agreement measured, and a classifier that matches the human label for source on at least 85% of a held-out 200 records; a published table of release sources by state and year.