Federal government to co-develop a government-wide AI tool with industry
Prime Minister Mark Carney's office says the plan is moving ahead: Ottawa will buy and build a government-wide AI system, and it will do it with help from Canada's AI sector. The tool will need to fit alongside the many AI pilots already running across departments.
The catch is clear. Internal AI leaders are warning against deploying tech for its own sake. The work starts with process redesign and better data, not a shiny new model.
What's actually being built
In the Nov. 4 budget, the government committed to develop "a made-in-Canada AI tool that can be deployed across the federal government." Shared Services Canada (SSC) will lead, supported by National Defence and the Communications Security Establishment (CSE). The stated aims: protect digital sovereignty, keep government data in Canada, and create opportunities for domestic vendors.
Industry will help define what gets built. As one PMO spokesperson put it, the tool "may not even exist yet," and the government may refine existing components with vendors to meet SSC's needs.
AI use is already widespread
Federal Chief Information Officer Dominic Rochon described AI use across government as "1,000 flowers blooming." Many programs aren't waiting for a centralized platform.
- Immigration and social benefits are triaging and processing files faster with AI.
- Fisheries and Oceans is spotting marine mammals in satellite and drone imagery.
- The RCMP is applying AI in human-trafficking and child-exploitation investigations.
- SSC built CANChat, a secure internal chatbot that knows basics like the April 30 tax deadline and does not train on user inputs.
Rochon said SSC is working on a registry to track which "flowers" are blooming and whether they are delivering measurable productivity.
Vendors in the mix, including Cohere
The government's relationship with Toronto-based Cohere is deepening, supported by financing for compute and an MoU to test its offerings. Cohere's North platform promises to build AI agents that execute digital tasks, not just summarize text. Whether Cohere will help determine needs for the new tool is still undecided, according to the PMO.
The internal warning: don't deploy AI for its own sake
Mark Schaan, a senior public servant working under the Minister of AI, cautioned against conference-fueled urgency. The right approach is to re-think the business process first: why the program exists, who it serves, and what outcomes matter. Then figure out where technology helps.
Preparation is non-negotiable. Schaan highlighted a financial firm that spent two years organizing its data before an AI rollout. Government faces the same reality: decades of siloed systems and inconsistent standards. Even procurement data suffers from quality and standardization issues, according to the federal procurement ombudsman.
For governance on automated decision systems, see the Treasury Board's Directive on Automated Decision-Making here. For security hardening, CSE's Top 10 IT Security Actions are here.
What this means for departments and agencies
Don't wait for the enterprise tool to fix foundational issues. Get your house in order now so you can plug in quickly when SSC rolls out standards and shared services.
- Clarify outcomes: define where AI can reduce backlogs, shorten cycle times, or improve service quality.
- Map processes: document steps, bottlenecks, decision points, and the data used at each step.
- Inventory data: owners, systems, sensitivity, retention, quality. Identify gaps and duplicates.
- Start data cleanup: standardize fields, reconcile IDs, de-duplicate records, and improve metadata.
- Set guardrails: privacy assessments, security controls, model access rules, and human-in-the-loop criteria.
- Measure baselines: current throughput, error rates, satisfaction, and costs. Define success metrics upfront.
- Pilot narrowly: one process, one outcome, time-boxed. Document results and lessons for the SSC registry.
- Engage the workforce: train users, adjust roles, and update SOPs. Plan for change management early.
- Coordinate with SSC: use approved platforms, meet security requirements, and prepare for shared components.
Procurement and implementation signals
SSC says it is already leveraging in-house and commercial AI to boost efficiency and will expand access across organizations. Expect vendor workshops, integration standards, and a tighter feedback loop on what works before broad rollout.
Formal details on the new tool and implementation will follow once the budget passes. In the meantime, departments should align pilots to anticipated enterprise patterns: secure chat interfaces, task automation, retrieval over departmental data, and auditability.
Key risks to manage
- Fragmentation and "shadow AI" if teams spin up tools without governance.
- Data quality issues that make models produce unreliable or inconsistent results.
- Vendor lock-in without portability plans, open standards, or exit strategies.
- Security exposure if model inputs or outputs leak sensitive information.
- Bias and fairness concerns if training data or prompts reflect historical inequities.
Practical next steps for program, policy, and IT leads
- Nominate an AI lead and cross-functional team (program, data, security, legal, procurement, HR).
- Publish a one-page use-case charter with outcomes, data sources, metrics, and risks.
- Complete a lightweight risk screen against the TBS Directive and your departmental privacy/security controls.
- Stand up a small, logged sandbox with test data and human review.
- Report pilot results to your CIO and feed into SSC's upcoming registry.
Upskilling your team
If your branch needs practical, role-based AI training to support pilots and adoption, explore courses by job for targeted options your team can apply immediately.
Your membership also unlocks:
