uniteTOlove.ca

Responsible Municipal AI Adoption — Playbook

How Toronto is actually using AI inside city government, and what real guardrails exist against misuse.

DRAFTThe playbookThe evidence file

What Toronto can do to make its own AI pilots visible and checked while no binding rule requires it to.

The honest bottom line

Toronto is already using AI — not "considering" it, actually piloting it in Public Health right now, including a predictive model that could shape which restaurants get inspected more and which get inspected less. None of this is required by any binding rule specific to Toronto: Ontario's Trustworthy AI Framework only binds the province's own ministries, and Bill 194, the one Ontario law that does reach the City, was criticized by the province's own privacy watchdog for leaving "all the critical rulemaking for future regulations," naming none of the actual principles, and giving residents "no clear or direct avenue" to complain about an AI-assisted decision — the Commissioner's own words, about her own province's law. Federally, the one serious attempt at a national AI law, AIDA, died when Parliament was prorogued in January 2025, and nobody has confirmed a replacement is actually coming. So the honest picture is this: Toronto is doing more, on its own initiative, than either the province or the federal government currently requires of it. That's a genuinely good sign about the people building the City's AI program. It is not the same thing as a guarantee, because good-faith internal initiative and a binding external check are different things, and right now Toronto has the first without the second.

---

a recommendation card — Extend Toronto's Existing Generative AI Guidance into a Published, City-Wide AI Use Registry

Card id: a recommendation card · Issue: ai-public-good-adoption · Backgrounder: our research file for that page · Trust: New load-bearing findings (NEW-2026-1) + carried-forward-LEDGER (CL-80031, CL-80032, CL-80033)

Problem

The City of Toronto's own interdivisional AI Working Group already lists an "AI registry" among its currently-underway deliverables, alongside an AI policy, generative AI guidelines, and an Algorithmic Impact Assessment framework [NEW-2026-1] — but this review did not locate a published, public-facing version of that registry, meaning residents and Council cannot currently see, in one place, which AI tools the City is using (beyond piecemeal disclosures like the two documented Toronto Public Health pilots) [CL-80032 “still being checked”, CL-80033 “still being checked”]. This sits inside a documented provincial accountability gap: Ontario's Trustworthy AI Framework's "No AI in secret" principle binds provincial ministries only, not Toronto, and Bill 194's municipal disclosure obligation defers its operative content to future regulation not yet in force [CL-0239 verified]. This card addresses only the visibility gap — it does not propose the City adopt or reject any specific AI tool, which remains a departmental operational decision.

Action

City Council directs that the AI Working Group's already-planned "AI registry" deliverable [NEW-2026-1] be published in a public-facing form (e.g., an open-data-style listing on toronto.ca), listing, at minimum, each AI system in active use or pilot by a City division or agency, the division responsible, the general purpose, and whether the tool was procured from a third-party vendor or built in-house — modelled on the transparency logic the EU AI Act applies to deploying bodies of high-risk AI systems [NEW-2026-9].

Jurisdiction split

Cost

Low (the registry itself is already a planned internal deliverable per the AI Working Group's own listed work [NEW-2026-1]; this card's incremental cost is the additional step of a public-facing publication layer, comparable in scale to the City's existing open-data-portal publication processes for other administrative datasets). No specific City open-data-portal per-dataset publication cost figure was located.

Funding path

Existing City Clerk's/Information & Technology division operating budget, using the same administrative channel already producing the AI Working Group's internal registry deliverable [NEW-2026-1]; no new funding source identified as necessary for a publication-layer addition to an already-planned artifact.

Who benefits, and how

Toronto residents and City Council, via visibility into which AI systems the City is actually using — visibility that does not currently exist in public-facing form in any source located in this review, despite the City's own governance framework and Ontario's Trustworthy AI Framework both naming transparency ("No AI in secret") as a stated goal [CL-0239]; civil-society and academic reviewers, via a single reference point rather than needing to track individual board reports.

Who bears the cost, and how

City taxpayers, via a marginal addition to existing open-data-portal administrative costs; City divisions using AI tools bear a modest new disclosure-compliance step within a process (the AI Working Group registry) they are already building for internal purposes.

Who benefits from the status quo

No beneficiary identified — the backgrounder's Cui Bono section found no ESTABLISHED or REPORTED finding naming a specific entity benefiting from the current lack of a public AI registry, a genuine coverage gap rather than an absence of possible beneficiaries in fact.

Financial ROI

Not yet estimable as a direct fiscal offset — this is a transparency/disclosure measure layered onto an already-planned internal deliverable, not a program with a modelled cost-avoidance or revenue case. No comparator found for this specific mechanism; the general order-of-magnitude comparator is the City's existing open-data-portal maintenance cost, not independently sourced.

Economic ROI

No source quantifies this — a public AI registry is a governance/transparency measure, not a program with a modelled local-growth, spending, or employment effect. A live search returned only general public-sector AI-adoption studies not scoped to a registry/disclosure mechanism specifically.

Social ROI

Directional: a published registry directly addresses the documented transparency gap between Ontario's own stated "No AI in secret" principle [CL-0239] and its current non-applicability to Toronto, and the Ontario IPC Commissioner's specific public criticism that Bill 194 lacks clear accountability mechanisms [NEW-2026-4] — though no source quantifies how much a published registry alone would change public trust or actual AI-governance outcomes. Confidence: low-medium — the direction is well-supported by the province's own stated principles; the magnitude of benefit is not quantified anywhere located in this review.

Environmental ROI

Genuinely environmentally neutral — a public-facing registry publication has no plausible emissions, land-use, water, waste, or resilience effect.

Evidence

Confidence & uncertainties

Medium confidence that publishing an already-planned internal deliverable is within existing municipal administrative authority (no new program is being proposed, only a publication step added to work already underway [NEW-2026-1]); low confidence on the registry's current internal completion status, since this review could not confirm how far along the AI Working Group's registry deliverable actually is. All NEW-2026-# citations are pending independent primary-source verification and formal formal registration.

Status

DRAFT — blocked on: confirming the AI Working Group registry deliverable's actual current completion status (not confirmed in this review); fairness and legal review; confirming no existing public-facing version already exists that this review simply did not locate.

---

a recommendation card — Require Human-in-the-Loop Disclosure for Any Toronto AI Pilot Touching Individual Service Eligibility or Risk Scoring

Card id: a recommendation card · Issue: ai-public-good-adoption · Backgrounder: our research file for that page · Trust: carried-forward (algorithmic-bias guardrail) + New load-bearing findings (NEW-2026-9)

Problem

Toronto Public Health's Food Safety team is preparing a proposal to pilot predictive AI modelling to "forecast the likelihood of future food safety infractions and optimize team resources" using DineSafe data, business information, 311 metrics, and environmental/wastewater surveillance data [CL-80033 “still being checked”] — a risk-scoring application of exactly the kind the underlying algorithmic-bias guardrail warns can "reproduce and amplify discrimination" if deployed without human-in-the-loop accountability and bias auditing. The underlying City board report already commits to a "health equity review... as a safeguard to prevent reflections of historical biases" for this specific pilot, so the bias-audit half of this concern is partially addressed on the record; what the same report does not describe is a human-in-the-loop review step before an enforcement action, or a defined appeal process for a flagged business — that narrower gap is this card's actual target. Neither Ontario's Trustworthy AI Framework (binding provincial agencies only) nor Bill 194 (whose operative regulatory content is not yet in force) currently imposes a binding human-in-the-loop or appeal-process requirement specifically on this kind of Toronto municipal risk-scoring pilot. This card addresses only the human-accountability design of this specific class of AI use — individual/business-level risk scoring or eligibility screening — not the City's AI governance program generally.

Action

City Council directs that any City division piloting or deploying an AI system that scores, ranks, or flags individual residents or businesses for a consequential outcome (e.g., inspection prioritization, benefits eligibility, code-enforcement targeting) must, before deployment, publish a plain-language disclosure — what the system scores, what data it uses, whether a human reviews flagged cases before an enforcement or eligibility action, and what appeal process exists — mirroring, at the municipal disclosure level, the EU AI Act's obligations on high-risk AI deploying bodies [NEW-2026-9].

Jurisdiction split

Cost

Low (a documentation and disclosure requirement layered onto AI pilots already being planned or piloted, such as the DineSafe predictive-modelling proposal [CL-80033]), anchored to the general comparator of the City's existing Algorithmic Impact Assessment framework deliverable already underway [NEW-2026-1], which demonstrates the City is already building comparable assessment infrastructure this card's requirement would use rather than duplicate.

Funding path

Existing City AI Working Group / divisional operating budgets, folded into the Algorithmic Impact Assessment framework already listed as an underway deliverable [NEW-2026-1]; no new funding source identified as necessary.

Who benefits, and how

Toronto residents and businesses subject to AI-assisted risk scoring or eligibility decisions (e.g., a food establishment flagged by the proposed DineSafe predictive model [CL-80033]), via a documented human-review step and a known appeal path — protections the underlying algorithmic-bias guardrail identifies as non-negotiable for consequential AI use but which are not currently guaranteed by any binding instrument reaching Toronto; City divisions, via a clearer internal standard to design pilots against before deployment rather than after a problem surfaces.

Who bears the cost, and how

City taxpayers, via modest additional documentation work within divisions already building AI pilots; no external payer class identified, since this is an internal governance-process requirement, not a fee or program cost imposed on residents or businesses.

Who benefits from the status quo

No beneficiary identified — same honest-empty-table finding as a recommendation card above; the backgrounder flags this as a coverage gap, not an absence of possible beneficiaries.

Financial ROI

Not yet estimable; this is a governance/disclosure requirement, not a program with a direct fiscal-offset case identified. No comparator found specific to human-in-the-loop disclosure requirements for municipal AI risk-scoring; the general comparator is the City's own already-underway Algorithmic Impact Assessment framework [NEW-2026-1], not independently costed.

Economic ROI

No source quantifies this — no local economic effect for a municipal AI-disclosure-standard requirement specifically has been modeled.

Social ROI

Directional: addresses a documented risk (algorithmic bias in consequential government decisions, harming "the racialized, poor, and marginalized" per the underlying guardrail's own framing) with a concrete, low-cost disclosure-and-human-review mechanism, though no source quantifies how much this specific disclosure requirement would reduce bias-driven harm in Toronto's own pilots. Confidence: low-medium — the direction is well-supported by the general algorithmic-bias literature; the specific magnitude of benefit in Toronto is not modelled.

Environmental ROI

Genuinely environmentally neutral — a documentation and disclosure requirement has no plausible emissions, land-use, water, waste, or resilience effect.

Evidence

Confidence & uncertainties

Medium confidence this is within existing municipal administrative authority (a disclosure/documentation standard, not new regulation of a third party); medium confidence the bias-audit component of this card's ask is already partly underway via the DineSafe pilot's own planned "health equity review" (per the board report, independently fetched this review); low confidence on whether the DineSafe predictive-modelling proposal or any other current Toronto AI pilot would, in its current design, already satisfy this card's human-in-the-loop review or appeal-process disclosure standard specifically, since neither element is described in the board report and this review did not obtain further pilot design documentation beyond it. All NEW-2026-# citations are pending independent primary-source verification and formal formal registration.

Status

DRAFT — blocked on: obtaining the DineSafe predictive-modelling proposal's actual design documentation to confirm whether a human-review step or appeal process is already planned alongside its stated health equity review; fairness and legal review; confirming this standard's relationship to the City's own in-progress Algorithmic Impact Assessment framework so the two are not duplicative.

---

Production record

Drafting record

Version: v2.0 (playbook conversion) · Original date: 2026-07-14 · Status: DRAFT · What this page draws on: formally registered claims (cited at recorded trust status), master briefing-carried-forward, NEW-2026-# source quotes. Author voice: The Unknown Soldier. Every factual premise traces to an existing formally registered claims row, the carried-forward master briefing, or a NEW-2026-# source quote. Per the costing bar (Q-06), all costs are order-of-magnitude ranges anchored to named comparators. Both cards stay strictly within documented municipal authority and do not propose that the City exercise powers (binding province-wide AI regulation, federal legislative action) it does not have. FIX-3 note (W1b cards audit, 2026-08-06, carried forward): a recommendation card's Action shortened to the template's one-action bar (was 103 words). No substance changed.

Playbook conversion (2026-08-11, Lane L3a): opened with "The honest bottom line" adapted from archive/dayone/ai-public-good-adoption.md (a recorded standing decision retired day-one memo, kept as history in archive/); each card's four-dimension "ROI schema v2" nested structure flattened to single tightened paragraphs per dimension, matching that page's recommendation cards's playbook shape; repeated "not yet estimable / genuine gap" boilerplate collapsed to one honest line each. All a formally registered claim and NEW tokens, figures, and comparators preserved unchanged.

About this file