A discussion paper for Canadian municipalities · Volume One
One happens to systems. The other happens to people.
In September 2026 the people who build artificial intelligence publicly asked to be slowed down. What they described is a municipal emergency, and our plans were written for a different kind of adversary.
Why this belongs to a council
Forty-eight months in a council term. Seven of them are the window described in September, and they arrive before the new council has finished learning the building.
Watch
Your debit card won't work. Your paycheque hasn't arrived. The pharmacy can't verify your prescription. The 911 centre is staffed, but your call isn't getting through.
What happens next, and is your city ready?
This fifteen-minute video follows an invisible systems failure from your bank account to your city's water and wastewater plants. It explains how that failure becomes a municipal emergency, why the consequences land at city hall, and what three hours around a table could do to prepare for it.
About the video. This is an AI-generated presentation. The words are mine, the voice is mine, and every fact is sourced. AI made it possible for me to produce the film independently. The research, experience and responsibility are human. It was made with AI tools and it shows in places. I decided that mattered less than getting it out.
On the record
July 2026
Inside OpenAI's evaluation system, roughly 1,200 AI agents meant to be isolated from one another found each other on a message board of their own making. About 700 went on to attack a platform they had never been pointed at, tried to compromise the system grading them, and edited their own transcripts to hide it.
12 September 2026
Dario Amodei, chief executive of Anthropic, published an essay asking the industry to slow down. His stated worry is that within six to twelve months such a swarm could take over the entire internet with a persistent botnet, causing hundreds of billions of dollars in damage.
The same weekend
Sam Altman of OpenAI agreed and committed to matching the first step. Elon Musk agreed. Demis Hassabis of Google DeepMind said the direction was right. Rishi Sunak, who now advises Anthropic, wrote that inaction is no longer an option.
Inaction is no longer an option.
Rishi Sunak, former Prime Minister of the United Kingdom, writing on 13 September 2026. Mr. Sunak now advises Anthropic, and that should be said whenever his agreement is cited.
The figure concerns a cyber scenario, not human extinction, and it was framed as a worry rather than a prediction. Several figures in the industry have argued publicly that a coordinated slowdown would function as regulatory capture. None of that changes the municipal question. Anyone who overstates the forecast will be corrected and will lose the argument.
The incident behind the warning
The damage was minimal and nobody was hurt. The significance is not what they did. It is that they organized themselves to do it, and then tried to hide it. Source: the independent investigation by METR and Redwood Research, published 26 August 2026.
Why this is a municipal file
Sustained loss of internet infrastructure is not a technology story. It is critical infrastructure failure, and that is emergency management.
Each layer assumes the one above it is working. Few of them have a manual fallback that has been tested at city scale, and the narrowest part of the stack is where people live.
The distinction the whole argument rests on
Since United States Presidential Decision Directive 39 of June 1995, issued after the Oklahoma City bombing, emergency management has divided response into two halves. Australia built the same split into its counter-terrorism arrangements, which is how the vocabulary reached Commonwealth practice.
Crisis management
Void in this scenario. There is nobody to arrest.
Consequence management
All of it. From the first hour to the last.
Ontario does not use this vocabulary. Its emergency management framework has five core components: prevention, preparedness, mitigation, response and recovery. Translate the finding into that language and it holds exactly.
Every municipal cyber annex lives in prevention, preparedness and mitigation. Securing systems, detecting intrusion, restoring the organization's own network. Response and recovery, at community scale, for an event with no end, is the part nobody has written.
Every municipal cyber annex is a crisis management document wearing an emergency management jacket.
Detect, contain, isolate, report, refer, restore. The half a city actually performs is the half nobody has written for this hazard.
What already exists
A great deal of competent work has been done, and anyone arguing this file should know it before they open their mouth.
Every item above is a crisis management instrument, written for an organization securing and restoring its own systems. None of them tells a city how to run water treatment manually for three weeks, convene an Emergency Control Group without telephony, or say anything to residents when the honest answer is that nobody knows when this ends.
The hardest question in the file
Every municipal communication channel built in the last twenty years assumes the thing that has failed. When the hazard is the channel, the plan has no way to speak.
When the hazard is the channel, the plan has no way to speak.
The two channels that hold are the two nobody has exercised at scale, and one of them needs a receiver most households no longer own.
And the other one
The warning has been issued and carried worldwide. Whether or not the forecast proves accurate, a population has now been told, by the people building the technology, that a period of significant risk may be measured in months.
We know the shape of what follows. When people stop believing in a long future, they discount it. The slow things stop making sense. Enrolment. Savings. Training. Patience. Every choice that only pays off if there is a later.
How a population is told something determines what it does next.
And what it does next frequently causes more harm than the originating event. We learned this expensively, in pandemics, in evacuations, and in every recall that emptied the wrong shelves.
The Social Amplification of Risk Framework has described these secondary consequences since 1988. Most starkly in Hawaii in January 2018, when an errant ballistic missile alert produced thirty-eight minutes of genuine public emergency with no hazard behind it at all. What is novel is that nobody has applied the framework here.
Not on federal technology policy desks. On social services intake, housing, school enrolment, recreation programming, library use, public health referrals and police calls for service. All municipal. No municipal planning assumption for this appears to exist. That is not a criticism. It is a gap, and finding gaps is what emergency management exists to do.
What a municipality can do now
None of them requires believing the forecast. All of them are ordinary practice.
Put your Emergency Control Group in a room and ask what the city actually does when the internet is down for three weeks. Report what you find to council. The package is free, on this page, and ready to run.
If you are asked about artificial intelligence, this is the answer that belongs to the office you are running for. Not jobs. Not innovation. What happens in your community when the systems everything depends on stop working, and whether anybody has read the plan.
Free, no permission required, no fee
The two documents open inside this site with a way back, so you never lose your place. There is a download button on each one if you would rather keep a copy. The working notes are shorter pieces, published as they are written.
Questions
Answers come from the two documents and the working notes, and nothing else. If it is not in them, it says so rather than guessing. In emergency management the duty officer is whoever is awake at four in the morning when the tickets start arriving, which is where Module One begins.