The Second Emergency All working notes Contact

Working Note 04

No One Is Required to Know

Working Note 03, "When System Failure Reaches the Kitchen Table," asked what happens between a machine being wrong and a person catching it. This note asks what law currently requires anyone to notice at all.Published 22 September 2026

Every question in Working Note 03 assumes somebody is watching. Somebody who knows which system has an AI agent inside it, somebody with the authority to stop it, somebody required to say so if they do not know. This note checked whether that somebody exists in Canadian law. It does not.

No statute currently requires a provincial or municipal essential service, or the vendor that runs it, to inventory where an AI agent has been given decision authority. Not to disclose it publicly. Not to assess the risk. Just to know it is there. That is not an opinion about weak enforcement. It is what is left when you read what actually exists.

You cannot govern a hazard nobody has been told to look for.

Three doors, and where each one stops

Three federal instruments are the ones most likely to be pointed to when this question comes up. Each one stops before it reaches a municipal water plant, a hospital record system or an ODSP case file.

InstrumentWhat it coversWhere it stops
The Artificial Intelligence and Data Act (AIDA)Would have regulated high-impact AI systems federallyDied on the order paper with Bill C-27 when Parliament was prorogued on 6 January 2025. The government has signalled it will not return in that form.
Bill C-26, the Critical Cyber Systems Protection ActCybersecurity obligations for six federally regulated sectors: telecommunications, interprovincial pipelines and power, nuclear energy, federally regulated transportation, banking, and clearing and settlementHas not received royal assent and is not in force. Progress froze at the same prorogation. Even in force, its six sectors do not include water, hospitals, income support or municipal government.
Treasury Board Directive on Automated Decision-MakingRequires federal institutions to complete an Algorithmic Impact Assessment before using an automated system in an administrative decision, with human review scaled to the impactBinds federal institutions only. It does not apply to Ontario, to a municipality, or to a vendor's system unless the federal government itself is the client.

Put plainly: the law that would have covered this is dead, the law that might one day cover part of it is not in force and was never going to reach provincial or municipal services, and the rule that already works is fenced to the federal government's own front door.

What Quebec has, and Ontario does not

One province has gone further than the federal government. Quebec's Law 25 requires an organization to inform a person when a decision about them is based exclusively on automated processing, to explain the factors and parameters that led to the decision on request, and to give the person a chance to have that decision reviewed.

Ontario has no equivalent for the provincial and municipal services this site is about. There is no requirement anywhere in Ontario law that a resident be told a system made an automated call about their benefit payment, their water, their care or their dispatch, let alone that a human review that decision on request.

Quebec's rule covers a decision about a person once it has already been made. Nothing in Ontario requires the service to even know, in advance, where that kind of decision is being made by a machine at all.

Who would need to be in the room

Working Note 03 asked who needs to be in the room. The honest answer is that the room has never been convened, because no statute tells any of its four members they have to sit down together.

Provincial

The provincial ministries that own the services

Nine separate ministries each hold a slice of the dependency map in Working Note 03, from Health and Children, Community and Social Services through to Municipal Affairs and Housing. None of them is required to ask whether an AI agent sits inside the systems they fund.

Federal, advisory only

The Canadian Centre for Cyber Security

It issues guidance on AI security and tracks the national threat picture, but it has no regulatory authority over a provincial or municipal system, and no mandate specific to AI agent behaviour as distinct from conventional cyber incidents.

Private sector

The vendors

The companies running case management software for ODSP and Ontario Works, hospital information systems, pharmacy networks, and payroll and payment platforms are under no obligation to disclose to their public-sector clients where they have handed decision authority to an agent, because no client contract or statute currently requires that disclosure.

Private sector

The AI labs

The companies building the systems that make this note necessary have no reporting relationship to any Canadian public-service regulator at all. Their own disclosures, like the ones cited in Working Note 03, are voluntary.

Each of the four could reasonably say the job belongs to one of the others. That is not evasion. It is an accurate description of a jurisdiction with no assignment.

The missing inventory

The federal government already has a working template for part of this. The Treasury Board's Algorithmic Impact Assessment requires a federal institution, before it uses an automated system in an administrative decision, to describe what the system does, how much impact it has on a person, and what human review is required as a result. It is not a perfect instrument and it was built for a narrower purpose than this note is raising. But it establishes a principle Ontario has not adopted for its own services: that an organization can be required to name, in advance, where a machine has been given a decision role.

The concrete ask that follows from this note is not a request to read another working note. It is a request for a specific mechanism: a provincial requirement, built on that same principle, that any system supporting an emergency-designated public service, whether the province built it or bought it from a vendor, be inventoried and assessed for where an AI agent holds decision authority, before that authority is granted rather than after something goes wrong. Emergency-designated is the operative word. This is not a proposal to regulate every use of AI in government. It is a proposal to treat AI decision authority inside 911, water, income support, hospital care and public payroll the way Ontario already treats other hazards to those same services, as something a hazard identification and risk assessment is required to name.

What the room would need to require

If that provincial requirement existed, it would need to answer the same practical questions Working Note 03 asked of individual organizations, but at the level of the sector, not the single municipality.

1. Inventory

An inventory obligation

Every vendor and public body operating an emergency-designated service must be able to state, on request, whether an AI agent holds decision authority inside that system and what that authority extends to.

2. Assessment

An impact assessment before authority is granted

Not after it is used, on the federal Algorithmic Impact Assessment model, scoped to emergency-designated services rather than every administrative decision.

3. Reporting

An incident-reporting duty

Specific to AI agent behaviour, distinct from a conventional cyber incident report, so a pattern across vendors and sectors can be seen by somebody before it is seen by the public.

4. Ownership

A named regulator

With the authority to receive that reporting. Right now no Ontario body has been assigned this file. Assigning it, even provisionally, is itself most of the work.

5. Recourse

A human review right

For the resident, on the Quebec model, when a decision affecting them was made or materially shaped by an automated system inside one of these services.

None of this requires Ottawa to pass AIDA, or to bring Bill C-26 back from prorogation, or to wait for a national AI law that may not arrive for years. It is provincial jurisdiction already, sitting inside the same authority that already governs hospitals, income support and municipal emergency planning in Ontario.

Recovery: the phase still to come

Working Note 03 covered what happens once a system has already failed or already acted, the machine being wrong and the moments after. That is not this note's job to repeat. This note's job sits upstream of it: what should have been known and required before the system was ever allowed to act in the first place. That is mitigation, in the language of the emergency-management cycle, and it has been genuinely missing.

One phase is still missing entirely, from this note and from the site. Recovery is what happens after a bad agent decision has already landed, the wrong benefit payment already issued, the wrong instruction already followed by a nurse or a dispatcher. In plain terms, fixing it once it has already gone wrong, not catching it before. Correction, reconciliation, and getting people to trust the system again afterward are a different job from stopping the error in the first place, and this site has not done that job yet. That is next.

What this means

Working Note 01 said the second emergency has no owner. This note can now say why, with citations rather than an impression. The federal instrument that would have addressed it is dead. The one that might have come closest was never going to reach this layer of government even if it passes. The one federal rule that does the actual work of naming a machine's decision authority in advance stops at the federal government's own door. Ontario has adopted none of it for the services this site is about.

The gap is not that nobody has noticed AI agents inside public services. It is that no one is required to.

That is a narrower problem than it sounds, and a more solvable one. It does not require a new federal law, a new international treaty or a resolution of what AI companies will or will not build. It requires Ontario to do to AI decision authority what it has already done to every other hazard in an emergency-designated service: name it, require it to be counted, and assign somebody to ask.

If a provincial or federal instrument already does this and I have missed it, or if you know of a jurisdiction that has built the inventory requirement described above, I would like to hear it.

Write to angela@lindow.ca.

Sources and verification notes

Verification note: this note describes the legislative and regulatory landscape as of 22 September 2026. Bill C-26 and AIDA's successor, if one is introduced, may change. Readers should check current status before relying on this note as a legal description rather than a working argument.

This note has been revised since it was published. Each change is recorded, with its date, in the site updates log.

All working notes Print or save as PDF Back to the documents

Free to use, adapt and reproduce. No permission required, no fee.

Working notes are dated when published. If one is revised, the change is recorded in the site updates log. Notes are never silently edited and never backdated.
TheSecondEmergency.com  ·  angela@lindow.ca