Working Note 09
Who Checks the Water?
Two Canadian warnings, one reassurance, and the gap between them. What the dates show, what the words say, and who has been told what to do.Published 1 October 2026
Over the past year, Canada's cyber security agency has published two warnings about attacks on the systems that run our water, and one reassurance about its own websites. Read side by side, with their dates, they tell a story about who gets told, who gets instructions, and who finds out last. This note sets it out in three parts.
Part One
The dates
1. What happened, and when
Here is the record, in order. Each line comes from a source listed at the end.
| When | What happened | Who was told |
|---|---|---|
| October 2025 | Someone interfered with the water pressure controls at a Canadian municipal water system, affecting service to a community. The municipality was not named. | The public, through a Cyber Centre alert reported on 30 October 2025 |
| 28 May and 9 June 2026 | An AI agent sent 899 requests to a search service run by Library and Archives Canada, including what researchers later called rudimentary hacking attempts. They do not appear to have worked. | Not made public for four months |
| July 2026 | OSFI, the federal regulator of banks and insurers, tells them to test what happens when AI fails and to keep manual fallbacks ready. | Banks and insurers |
| 30 July 2026 | The FBI and the US Environmental Protection Agency report attacks on the controls of American water utilities. | Water operators |
| 19 August 2026 | Five US agencies warn that AI is making these attacks faster and easier. | System operators |
| By 14 September 2026 | Canada's Cyber Centre warns that attackers are using AI against the controls of water, energy, food and other systems, and that future incidents in Canada will likely be more severe. It repeats the October 2025 water incident. | System owners and operators |
| 28 September 2026 | Transluce, an independent research lab, tells the Government of Canada about the Library and Archives Canada attempts. | The government, privately |
| 29 September 2026 | The Cyber Centre says it is aware of suspicious activity, including suspected AI agent activity, targeting public websites, and that there is "no indication that government systems have been compromised at this time." | The public |
| 30 September 2026 | Transluce publishes its findings. News reports follow. | The public |
The attempts happened in May and June. The government was told on 28 September. It answered the public the next day.
2. What the dates show
The reassurance of 29 September came one day after an outside research lab told the government what it had found, and one day before that lab told everyone else. The researchers found the evidence in pages saved by Arquivo.pt, a public web archive based in Portugal.
The researchers were careful. They said the attempts do not appear to have succeeded, and they did not firmly attribute them to any company. The Cyber Centre was careful too. It said it had no indication of a compromise "at this time" and that its assessment was continuing.
What the dates do show is a pattern. In Working Note 08, two AI incidents each took about three months to reach the public. This is a third, and it happened here.
Three incidents. Three gaps of three months or more. The third one was in Canada.
Part Two
The words
3. Two statements about two different things
The warning, by 14 September
AboutThe controls that run water, energy, food and other systems.
SaysAttackers "changed device settings, disabled alarms, and locked operators out of their systems." Canada has not yet had a high-impact incident, and future ones will likely be more severe.
Written forThe people who own and operate those systems.
The reassurance, 29 September
AboutPublic websites, "such as the Government of Canada."
Says"There is no indication that government systems have been compromised at this time." The assessment continues.
Written forThe public.
Neither statement takes back the other. The warning about water is still on the Cyber Centre's site. The reassurance is about websites. Each may be accurate. Read quickly, though, the second sounds like an all clear, and it was the one written for the public.
4. What "government systems" covers
A federal agency can speak for federal systems. The water from your tap is a different matter. In Ontario it is run by your municipality, or by a utility your municipality owns or hires. A statement from Ottawa about its own websites cannot tell you whether anyone has checked the controls at your water plant, and it does not claim to.
"No indication of compromise at this time" means that, when the statement was written, the agency had found no evidence of a break-in. That is an honest, limited sentence, and it deserves to be heard as one. Nothing in it says anyone checked the water.
A statement about government websites is not, by itself, an answer about the water coming out of our taps.
5. What we were not told
The October 2025 water incident has been public for almost a year. The municipality has never been named. We do not know whether its residents were told, how the attacker got in, how long it took to notice, or what stopped it from getting worse. Some technical details deserve protection. The basic facts of a public service being tampered with are a different matter: residents can be told that much without anyone being put at risk.
Part Three
The instructions
6. Who has been told what to do
Banks
Told in July to test AI failure and outage scenarios, keep manual fallbacks, and make suppliers say "if and how they are using AI to deliver services."
A binding rule on AI and other models takes effect on 1 May 2027.
Water systems
Warned that attacks using AI are increasing and will likely get worse.
Advised to keep their controls off the open internet. An alert, not a rule.
Households
Nothing.
No plan, no notice, and no one required to tell them.
The federal bank regulator moved first, and it moved well. It assumed AI would fail, and told the banks to have a way to keep working by hand when it does. That is ordinary emergency management, and it is exactly what is missing for the water.
The banks have been told to plan for AI failure. The people who drink the water have not been told anything.
7. Why an AI helper changes the picture
An AI helper, sometimes called an AI agent, is a program that can act on its own: it can log in, search, change a setting or send a message without a person approving each step. OSFI put it plainly. These systems "can act with limited human oversight."
That is useful when the instruction is right. When the instruction is wrong, or has been slipped in by someone else, the helper carries it out just the same, and quickly. The attempts on Library and Archives Canada were automated: 899 requests, sent by a machine, found months later by researchers abroad.
Imagine if
It is a cold Tuesday in February. A company hired to tune a city's pumps uses an AI helper to adjust the schedule overnight. A hidden instruction in an email it reads tells it to change a pressure setting. It does. By morning, three apartment towers on the same street have a trickle from the tap.
The building manager calls the city. The city calls the contractor. Residents see a notice that says "maintenance." Nobody is required to tell them what happened, whether it could happen again, or what to do in the meantime.
An AI helper carries out the instruction it is given, right or wrong. The first sign may be the tap.
8. Questions for the people responsible
For the Canadian Centre for Cyber Security: when did you first learn of the activity on the Library and Archives Canada website? Did your 29 September assessment cover any provincial or municipal systems, including water? Were the residents served by the October 2025 water system told what happened?
For the Province of Ontario: which ministry is responsible for the cyber security of municipal water controls, and what does it require of them?
For municipalities: can your water controls be reached from the internet? If they stopped working, or started taking the wrong instructions, how would you run the system by hand, and how would you tell residents?
For all three: if Canada's bank regulator can require manual fallbacks for AI failure, who will require the same for water?
9. What a household can do now
You cannot inspect a water plant. You can decide in advance what your household does if the water stops, or comes out wrong. The Plan sets out five first steps for the day it happens, two sheets to fill in by hand, and what to check once service returns. It is free, in English and French.
Smaller dated events go to On the Record, one line and one source each.
Sources
- Canadian Centre for Cyber Security. "Statement regarding reported activity targeting Government of Canada websites." 29 September 2026. cyber.gc.ca
- Canadian Centre for Cyber Security. "Cyber threat actors use artificial intelligence in an active global campaign to disrupt internet-exposed programmable logic controllers." Page modified 14 September 2026; first publication date not confirmed. cyber.gc.ca
- Transluce. "AI Agents Targeted U.S. and Canadian Government Websites." 30 September 2026. transluce.org
- Global News. "An AI agent tried to hack Canadian government website, researchers say." 1 October 2026. globalnews.ca
- CP24 / The Canadian Press. "Canada's Cyber Centre warns of hacktivists targeting water, energy systems." 30 October 2025. cp24.com
- Office of the Superintendent of Financial Institutions. "Generative and Agentic Artificial Intelligence: Implications for Technology, Cyber Security, and Operational Resilience." July 2026. osfi-bsif.gc.ca. OSFI. "Guideline E-23: Model Risk Management (2027)." Effective 1 May 2027. osfi-bsif.gc.ca
- FBI and EPA. "Malicious cyber actors targeting water and wastewater sector internet-facing programmable logic controllers, causing operational disruptions." 30 July 2026. fbi.gov
- CyberScoop. "AI-fueled attacks pose 'active threat' to water, other sectors, U.S. agencies warn." Advisory AA26-231A, 19 August 2026. cyberscoop.com
Verification note: every date above is as shown in the sources. An earlier note, Working Note 07, gave the Cyber Centre's industrial-controls warning as dated 8 September 2026. The page shows it was modified on 14 September 2026, and its first publication date has not been confirmed. Note 07 was corrected on 1 October 2026, and the change is recorded in the site updates.
All working notes On the Record Print or save as PDF Back to the documents