Policy Analysis - April 2026

Building Canada's Early Warning and Response System for AI Job Displacement

A proposal for how Canada can detect AI-driven labour displacement before it reaches crisis scale, and how Mila can shape the national response.

↓ Download Executive Summary

Full report available on request: rodmoshtagi@hks.harvard.edu

Report cover: Building Canada's Early Warning and Response System for AI Job Displacement
The Problem
$25M
allocated to TechStat in Budget 2025 to measure AI's impact on the Canadian labour force. As of April 2026, no formal tracking framework is operational.
66,705
early-career Canadian workers aged 20 to 24 across seven occupations where U.S. evidence suggests AI-driven displacement is emerging. These workers sit at the intersection of high AI exposure and low labour market experience.
Customer service reps
29,605
Financial service reps
9,400
Accounting clerks
8,815
Software developers
8,005
Software engineers
5,805
Web developers
4,770
Travel counsellors
305
43.4%
of these at-risk early-career workers are concentrated in Ontario. Quebec holds another 23.6%. Together, two provinces account for two-thirds of Canada's most exposed young workers.
BC 12.5% AB 7.0% SK MB 5.1% ON 43.4% QC 23.6% NB NS PE NL
The System

Four Data Streams, Layered by Dependency

The early warning system should be built on domestic data first, with frontier lab telemetry treated as a plug-in upgrade.

Highest value - longest lead time
AI Usage Statistics from Frontier Labs
Forward-looking signal: shows where automation pressure is building. Requires voluntary data-sharing agreements. Earliest realistic delivery: early 2027. Open-weight models create a 20-30% blind spot.
↓ Plugs into ↓
Requires partnership
High-Frequency Payroll Data
Confirms actual displacement: hiring freezes and separations in near-real-time. ADP Canada and Dayforce cover roughly 40% of the workforce combined; 80% with additional providers.
↓ Validates ↓
Within government control
System Backbone
LFS digital platform supplement (needs quarterly frequency), dynamic task exposure metrics (no external dependencies), and firm-level AI adoption surveys (CSBC sunsets August 2026).
TechStat should build this system to operate without frontier lab data, and create standardized plug-in points that labs can connect to on their own terms.
Key Findings

What the Analysis Reveals

Six findings from the analysis. Together they describe the emerging displacement signal and the gaps in Canada's ability to detect and respond to it.

01
At-Risk Early-Career Workers
Roughly 66,705 workers aged 20 to 24 sit inside seven occupations where U.S. evidence suggests AI-driven displacement is emerging. Customer service representatives, software developers, and financial service reps account for 78% of the cohort. Two-thirds are concentrated in Ontario and Quebec.
02
The Coverage Gap
Open-weight models account for an estimated 20 to 30% of all AI usage, and when users run these models on their own hardware, no usage statistics are generated. Any system anchored solely on frontier lab data is structurally blind to this segment.
03
The Single-Lab Skew
Coding-task share differs by a factor of nine between Claude (37%) and ChatGPT (4.2%). Relying on one provider's data as ground truth creates a distorted picture of what work AI is absorbing across the economy.
04
Canada's Survey Gaps
Labour Force Survey cells are too small to detect cohort-level shocks, and the Census operates on a five-year cycle. Payroll data closes this gap.
05
CERB's Warning
$4.6 billion in confirmed overpayments traced to the inability to verify employment status in real time. The Auditor General explicitly recommended developing a real-time payroll system.
06
The EI Eligibility Gap
Employment Insurance requires prior insurable hours. Young workers entering the labour market for the first time fall outside it.

Without an early warning system, the next labour disruption will follow the CERB pattern: detected late, designed under pressure, deployed without verification, and coordinated with provinces after the damage is done.

The Data

Evidence Behind the Design

Coverage Gap: What Frontier Labs Can See
70-80% visible
Visible to frontier labs
Open-weight blind spot
Sources: Menlo Ventures (Nov 2025), a16z/OpenRouter (Dec 2025).
Single-Lab Skew: Coding Task Share
Claude
37%
ChatGPT
4.2%
A 9x divergence. An Anthropic-only signal would overstate software automation and understate shifts in customer service and healthcare.
Sources: Anthropic Economic Index (Feb 2025), Chatterji et al. NBER 34255 (Sep 2025).
The Cost of Not Knowing in Real Time
$4.6B
in confirmed overpayments when Canada deployed emergency benefits without real-time employment verification. At least $27.4 billion more was flagged for further investigation. The Auditor General recommended building a real-time payroll system.
Source: Office of the Auditor General, 2022 audit of COVID-19 benefits.
Recommendations

Six Recommendations for Mila and the Government of Canada

For Mila
1Brief TechStat on the indicator framework in this report and advocate for the data streams it will need, engaging in Q2 2026 before TechStat's first-year work plan is finalized.
2Convene a Canadian AI labour displacement scenario exercise in October or November 2026 that produces a scenario-to-policy matrix mapping plausible displacement scenarios to candidate policy responses.
3Consult provinces through bilateral outreach to Ontario and Quebec and a joint technical briefing with the Labour Market Information Council, and communicate Canada's data requirements to the international AI usage data standard effort.
For the Government of Canada
4TechStat should secure three domestic data streams in 2026: an LFS digital platform supplement at quarterly frequency, absorption of the CSBC AI adoption module before its August 2026 sunset, and voluntary data-sharing partnerships with ADP Canada and Dayforce.
5TechStat and ESDC's Labour Market Information Directorate should begin quarterly calibration sessions in mid-2026 to develop shared signal interpretation, escalation thresholds, and signal-to-response mapping.
6ESDC should table two sequenced items at the FLMM's Changing Nature of Work and Skills Working Group: an awareness agenda item in late 2026 and a structured working session on means-testing treatment of future AI displacement benefits in mid-2027.

The full report includes detailed implementation plans, timelines, and named counterparts for each recommendation.

Q2 2026
Mila briefs TechStat
Aug 2026
CSBC sunsets
Oct 2026
Scenario exercise
Q1 2027
First payroll data
Early 2027
Lab standardization
Mid 2027
FLMM means-testing

About the Author

Rod Moshtagi

Rod Moshtagi is an MPP candidate at the Harvard Kennedy School and a former consultant at PwC Canada, where he served as a strategic advisor to Employment and Social Development Canada's Benefits Delivery Modernization program on data governance, data privacy, and analytics.

Harvard Kennedy School
Prepared For
Mila - Quebec Artificial Intelligence Institute