That the deciding control was a person who could still run the plant by hand. Where crews noticed quickly and knew the manual procedure, the attacks became inconveniences. That capability decays unless it is practiced, and every new layer of automation makes it easier to lose.
On July 26 and 27, 2026, more than 30 Minnesota community water systems were targeted in what Minnesota IT Services described as a coordinated cyberattack. On July 30 the FBI confirmed that water and wastewater utilities in seven states had been hit. By mid-August, as the Christian Science Monitor reported, no serious damage had been done. The reason it was small is the part worth studying.
What the attackers did, and what stopped them
According to the Monitor's reporting, the attackers reached internet-facing programmable logic controllers and changed their IP addresses and passwords, locking utilities out of their own equipment. The method was not sophisticated. What it removed was the operator's normal view of, and control over, the plant.
The local accounts show what replaced it. In Braham, the city reported at 9:34 a.m. on Monday, July 27 that its water plant was offline, and public works crews had it back in under two hours with no reported effect on water quality. In Plymouth, the problem hit cellular-connected equipment at two water towers and several lift stations, and crews kept the system running manually. South St. Paul and Maple Plain reported that staff maintained normal operations even though some automated controls were affected, although Maple Plain also declared a local emergency. Karleen Kos, CEO of the Minnesota Municipal Utilities Association, told the Monitor that some operators lost visibility into their systems, some switched to manual controls, and some shut equipment down, rebooted it and corrected the anomalies they found.
Cynthia Finley of the National Association of Clean Water Agencies summarized it this way: "Utilities were able to notice the attack quickly, take action quickly, and there were no dire consequences." She added that the sector is "concerned about more sophisticated cyberattacks."
The Human-System Unit when the feedback channel breaks
Anthromekagogy analyzes a facility as a Human-System Unit: the operator, the constructed system, the task environment and the feedback channel between them. In a water plant, the feedback channel is the controller, the network and the display that tell an operator what the pumps, valves and tanks are doing. The July attacks cut that channel. What remained was the operator's own knowledge of the plant and the physical means to act on it.
That is the framework's reversibility constraint under live conditions: people must be able to function without the system. In July, at small utilities that had kept manual procedures alive, reversibility held. It did not hold because of any cyber control. It held because a public works crew knew where the valves were and what normal looked like.
The scale of the sector makes this more than a local story. The Monitor notes, citing a 2022 National Conference of State Legislatures report, that nearly 90 percent of U.S. public water utilities serve fewer than 10,000 people. Many cannot fund dedicated security staff. For those systems, an operator who can run the plant by hand is not a backup to the security program. It is the security program's last and most important layer.
Why the capability decays
The framework names the mechanism that erodes this capability. The skill atrophy loop runs from less practice, to weaker competence, to more reliance on the system, to further decline. It proposes a Skill Retention Half-Life to measure it: the time for an independently executable skill to decay by half under reliance on a system. Neither construct has been measured for water operators, and this article does not claim a value. The point is that the quantity exists and that nobody tracks it.
Every automation upgrade that works well shortens the time operators spend running equipment by hand. Remote monitoring replaces site visits. Automated dosing replaces manual adjustment. The next layer is already arriving: the joint principles for integrating AI into operational technology published by CISA and international partners in December 2025 call for keeping humans in the loop and for AI-enabled systems that fail gracefully without disrupting critical operations. Those are the right principles. They depend on an operator who is still capable of taking over, and that operator is produced by practice, not by policy.
This is the pattern the framework calls automation drift: the gradual removal of human agency through efficiency gains that are each reasonable on their own. No single upgrade removes the manual fallback. The sum of them can.
What the guidance now asks for
Government guidance is moving toward the same conclusion. On July 28, the Australian Signals Directorate published its CI Fortify advice on isolating vital systems, developed with international partners. It warns that isolating systems "will trigger manual processes," asks operators to define trigger criteria for each isolation step in advance, to test the isolation of all vital systems periodically, and to keep a secure offline hard copy of the plan. In other words: decide in calm conditions what people will do when the network is gone, and rehearse it.
The timing also matters. As IOActive pointed out on August 10, the federal advisory on Iranian-affiliated targeting of internet-facing controllers, AA26-097A, was first published in April and updated on July 22, four days before the Minnesota activity. No connection between the two has been established. But the warnings were available and free, and the constraint at small utilities was capacity, not awareness. When the preventive layer is thin, the human fallback carries more of the load.
Questions owners should be able to answer
- Which processes can we run by hand today, and for how long? Name them, with the staffing each requires.
- When did each crew last do it? A procedure nobody has performed in two years is a document, not a capability.
- What tells an operator that the display is wrong? Local gauges, tank levels read in person and field checks are part of the feedback channel too.
- Who decides to isolate, and on what trigger? CI Fortify asks for this in advance, in writing, on paper.
- Does each new automation or AI tool shrink the manual envelope? Ask before procurement, not after an incident.
The July 2026 attacks were contained by people, not by technology. Treat manual operation as a measured capability with a practice schedule, and test whether each new layer of automation leaves crews more capable or more dependent.
Sources
- Gass, H., Cyberattacks on water facilities test US defenses, reveal gaps, Christian Science Monitor (Aug. 18, 2026)
- CBS Minnesota, “Coordinated cyberattack” targeted 30-plus Minnesota water systems in 48 hours, MNIT says (July 28, 2026)
- IOActive, When the Advisory Arrives First: Minnesota’s Water Utilities and the Limits of Warning (Aug. 10, 2026)
- Australian Signals Directorate, CI Fortify: Advice for isolating vital systems (July 28, 2026)
- CISA and partners, Principles for the Secure Integration of Artificial Intelligence in Operational Technology (Dec. 2025)
- Kristof, R. (2026). Anthromekagogy: A Research Framework for Human-System Co-Evolution. Working paper.
Cite this article: Authentic Intelligence Research, “What Did the 2026 Water Attacks Reveal About Manual Fallback?” Authentic Intelligence, October 12, 2026, https://authenticint.com/articles/manual-fallback-after-the-2026-water-attacks.html.