AI Strategy for the Federal Public Service 2025-2027: Priority areas

The AI Centre of Expertise (AI CoE) will complement services offered by existing accelerators and centres, focusing on project support, knowledge sharing, and strategic guidance to support AI adoption and experimentation. It will provide project guidance, advise on common processes, and share best practice, experience, knowledge and code. It will also encourage interdepartmental or whole-of-government collaboration on solutions for common needs. The AI CoE will help business teams to:

• Identify high-value use cases for AI integration. This may include helping departments evaluate factors like task and data suitability; infrastructure, model, and data requirements and limitations; cost/benefit analysis; privacy, accessibility, environmental, and equity considerations; risk identification, evaluation, and mitigation; workforce impacts; and how to evaluate project success against objectives.
• Support data readiness: This would include supporting the implementation of the 2023-2026 Data Strategy for the Federal Public Service as a critical enabler for AI adoption. The AI CoE could advise departments on preparing their data for AI use or training and identifying Government of Canada datasets that could be combined to enable AI.
• Support with procurement, governance, assessments, review processes, and other requirements. This would include providing guidance on options for procurement; algorithmic, accessibility, language, privacy, and environmental impact assessments; transparency, reporting, ethics, bias, peer review, GBA Plus analysis, and Indigenous Data Sovereignty requirements; securing AI and understanding risks; and processes for explanation and recourse. It will also include providing input into the development of a governance framework (see Priority 2) and the creation of standard language to inform vendors of their role in responsible AI adoption.

In addition to supporting project teams, the AI CoE will act as:

• A convenor for government-wide knowledge sharing. It will share successful initiatives to prevent duplication of effort, maximize return on investment, and promote awareness and adoption of existing approved solutions. This will include sharing government-developed code and solutions, project documentation, information, policy, and instruments; soliciting feedback, use cases, and lessons learned from departments; and encouraging the scaling of successful projects. It will also work with the Canada School of Public Service (CSPS) to identify and promote training priorities and provide input into new policy or procurement vehicles.
• An accelerator for AI solutions to meet common Government of Canada needs. Working with departments, the AI CoE will identify potential solutions that meet government-wide needs that could be broadly scaled and support their development.
• A monitor of progress on AI adoption. The AI CoE will develop frameworks for progress and impact of Government of Canada AI adoption that can be used to benchmark, monitor, and report on the progress and compliance with policy and standards.

The Government of Canada and departments will ensure the provision of common infrastructure to enable AI adoption and will promote the adoption of existing approved enterprise solutions. This includes:

• high-performance computing (HPC) and cloud infrastructure that is available, secure, and scalable to meet the demands of AI projects. 
• common data and information management systems and practices to support sharing, scaling, and agentic capabilities across Government of Canada platforms.
• access to approved models, services, and application programming interfaces (APIs) within common Government of Canada cloud platforms to build or deploy systems.
• access to common AI solutions and capabilities in vendor solutions through a standard Government of Canada platform service.

Where possible, they will support Canadian suppliers and vendors in order to promote the growth of the Canadian AI industry and ensure secure sovereign infrastructure and solutions.

This project will be either a new initiative or one currently at pilot stage that would meet an enterprise-wide need and could be readily scaled. Its development, testing, and delivery will be undertaken collaboratively by the AI CoE and lead department. It will serve as:

• A test case to identify barriers and project teams’ needs for support during development and obstacles to subsequent customization and scaling.
• A pilot for the development of government-wide governance processes (see Priority 2 ).
• An opportunity to develop templates and examples of project documentation.
• A demonstration of the value of AI use cases.

Drawing on international and Government of Canada best practice, the Government of Canada will establish common governance and risk management frameworks for the AI lifecycle to provide clear guidance to AI project teams. These frameworks will balance risk with the pace required for innovation; be scaled to system risk, sensitivity, and impact; and will be designed to adapt as technology changes. They will address potential risks associated with AI use, such as data privacy and security, bias detection and mitigation, model interpretability and explainability, environmental impact, and human involvement. The frameworks will incorporate Canadian requirements, such as the need to advance Indigenous Data Sovereignty and provide technologies in both official languages. They will also lay out the security, trust, and reliability controls necessary to ensure the continued maintenance and resilience of these systems, including system shutdown measures to respond to emergencies. This work will support ongoing efforts to harmonize data standards, increase interoperability, and support other enablers.

In addition, it will include a government-wide governance structure responsible for implementing the governance framework. This will include an AI ethics review board responsible for providing guidance on responsible AI and for evaluating higher risk and impact projects at key stages of the project lifecycle. The governance structure will make use of existing Government of Canada governance bodies and organizations such as the Canada AI Safety Institute wherever possible. In addition to providing project governance, these bodies and organizations will help identify high-value AI use cases, regularly reprioritize the AI Strategy's actions, and prioritize and oversee the AI Strategy's implementation.

The Government of Canada will review and propose changes to instruments that create unnecessary obstacles to AI adoption. It will fill identified policy gaps, such as clarifying the responsibilities of chief information, data and privacy officers for AI adoption, and the acceptable use of AI by outside organizations in their interactions with the Government of Canada. It will also address legal and policy ambiguities related to privacy and training data and the application of the national security exemption. The Government of Canada will update procurement policies, instruments and processes to make them more responsive to pace and requirements of AI procurement. It will also consider ways in which the Government of Canada can better align internal AI policy with the Pan Canadian AI Strategy to support the Canadian AI sector, with the United Nations Declaration Act commitments to Indigenous Data Sovereignty, and with international treaties, legislation and norms.

This work will lead to the development of clear and concrete commitments to revise specific instruments within a set timeline, and an agile process to review and update policy, guidance, tools and resources to respond to technological, legislative and social change. The Government of Canada will also review the policy landscape for opportunities to synthesize or interpret existing policy to increase usability.

The Government of Canada will optimize its AI adoption by embedding a “think AI” approach to its policies, programs and services. The intent is not to adopt AI at all costs or in contexts where its use would not be responsible or useful. Rather, the goal is to challenge departments to identify key business problems that could be transformed using AI, consider AI options before defaulting to traditional IT or HR approaches, and make planned investments in AI and its enablers.

To support this, the Government of Canada will require departments to:

• Identify three areas, programs or services with business problems that have a high potential to be solved using AI.
• Consider solutions and resourcing requirements for AI and its enablers at the outset of initiatives. This will include changes to Treasury Board submissions, Memoranda to Cabinet, and budget and off-cycle funding proposals to incorporate data, AI, compute, and other relevant requirements.
• Prioritize AI infrastructure and secure adoption in departmental integrated IT planning processes.
• Develop their own AI strategies to ensure alignment and effective use of resources.

Building on existing offerings from CSPS, the Government of Canada will develop a training plan for existing public servants. The developed training plan will be evergreen to match the pace of AI development. Through this plan, the Government of Canada will consider ways that AI may reshape the workforce, working with employees and their bargaining agents to prepare public servants for this change through retraining.

The training plan will incorporate both general training and training tailored to specific personas with varying roles, levels and responsibilities. The general training will be directed at increasing understanding of and confidence in using AI, including embedded AI capabilities; developing skills associated with responsible, secure, and effective use, including effective prompt engineering; and establishing leadership programs to achieve a culture that promotes AI adoption. The tailored training will address both more specific and advanced technical skills, and the behavioural skills needed for successful adoption, such risk identification and management, and effective leadership of AI projects and teams.

The Government of Canada will benchmark both talent requirements for AI and its employees’ existing AI knowledge and skills across the enterprise. These benchmarks will enable the development of learner personas with accompanying training plans and identify employees who could be offered further training.

Since not all talent needs can be met through training, the Government of Canada will need to develop a plan to recruit and retain talent. This plan will explore obstacles to recruitment and retention and ways to establish flexible data science career pathways for AI practitioners. It will also consider ways to expand interchanges, apprenticeships, co-op programs, and partnerships with AI institutes and research centres to create a talent pipeline. It will consider ways to make efficient use of AI talent through flexible assignments, and competitions and challenges to meet some project-based needs.

The Government of Canada will continue to strengthen and clarify accountabilities for AI use in policy and in notifications about AI use. These will include new requirements and standard language for the disclosure of AI use, explanations of how an AI system reaches a decision, and information about rights and protections, including how to seek explanations or recourse and how to report problems. Lastly, the Government of Canada will identify those AI capabilities that it will not pursue.

As part of these efforts, the Government of Canada will prioritize the establishment of a public register of its AI systems. The register will include information about all AI systems that fall within a defined scope, drawing wherever possible on information already collected through Algorithmic Impact Assessments and Personal Information Banks. The scope of the register will be defined and publicized on the register itself and will follow best practice for registers in other jurisdictions, which exclude embedded AI systems and any systems covered by specific policy prohibiting publication. The register will include information about what and how data is being used, how it was trained, and what quality assurance and privacy and security measures are in place. Where information about systems cannot be published, the Government of Canada will develop and publish alternative oversight processes.

To ensure good stewardship of public resources, the Government of Canada will develop metrics and performance indicators to demonstrate the impact and value of AI initiatives to those we serve. This will track both the cost-effectiveness and cost-efficiency of AI projects through a range of financial and non-monetary metrics, including costs of implementation, efficiencies and cost savings achieved, service improvements, increased client satisfaction, or increased program uptake.

The Government of Canada will also develop a framework to track AI adoption. This framework will incorporate metrics for the deployment of AI solutions overall, but also for metrics such as collaboration, sharing, or scaling of solutions to avoid duplication of effort, maximize investment, and reduce costs. The metrics will also track departmental investments in the enablers of AI, including data, infrastructure, and talent.

In keeping with its commitments to meaningful public engagement through the Directive on Open Government, and with the duty to consult, the Government of Canada commits to early and meaningful public and stakeholder engagement on AI initiatives of significant public interest or concern. This will include targeted engagement of communities that face greater impacts, risks or barriers from AI systems and union and employee engagement on workforce impacts. It will also include client participation in system design to ensure that AI systems do not create or perpetuate barriers to access for clients. The Government of Canada will also provide mechanisms for ongoing public feedback and questions on AI used by the federal government.