Checks and balances: how to deploy responsible AI in government

Image by rawpixel.com on Magnific

Key to governments using artificial intelligence effectively is designing and implementing the appropriate guardrails so that its outputs are safe, fair and explainable. During this webinar, experts in Canada set out how to drive the adoption of AI in government, with responsible use front of mind

Governments working to implement artificial intelligence to reduce administrative burden and improve public service delivery are doing so amidst public concern about the risks of the technology and against the backdrop of a general decline in trust in public institutions.

So, if AI is to be embraced and drive successful outcomes, responsible deployment and oversight is key.

At a Global Government Forum webinar supported by knowledge partner SAS, and for a Canadian audience, public and private sector leaders in Canada discussed how to implement responsible AI, covering topics such as data privacy, training, securing the buy-in of staff, and striking a balance between checks and balances and the freedom to innovate.  

Well-managed data the foundation of responsible AI

The conversation began with the underpinning of all AI systems: data.

Alexandra Dykes is director of external data services within the privacy and responsible data team at the Treasury Board of Canada Secretariat (TBS).

“What everyone – Canadians in this case – forget about is that data drives AI, and well-managed data is really a foundation of being able to implement AI, so that we can trust the results,” she said, adding that there is a misconception among Canadian citizens about the extent of government’s use of their data.

They tend to believe that because they give their data to government, government should know “everything there is to know about them”, she said. “That is indeed not the case. We take data protection incredibly seriously”.

The country’s privacy act provides guardrails around what can be shared between programmes and between departments, she noted, and the federal government is planning to modernise both this and its information act so they are “current in today’s reality”.

“I think something that is really important for people to remember is that safeguarding of personal information, as well as our information and data holdings, is key. There’s a lot of legislation that supports that.”

Dykes also said that she sees room for improvement in the way government makes citizens aware of the pace and scale of its AI adoption. Citizens could be forgiven for thinking government is already using it in every corner of its operations, leading to nervousness, she explained, but it’s actually much less prevalent than people realise.

And where it is being used, there are safeguarding policies in place and more being developed as adoption accelerates, as well as tools and guidance that the TBS develops and disseminates to departments.

“The Government of Canada is supporting responsible [AI] adoption through controlled tool use and departmental oversight and workforce enablement, so we have tools that are developed and rolled out to departments that really have very clear from-the-centre guidelines on how that should be done,” Dykes said.

Read more: Competition calls for Canadian public servants to submit ideas for data and AI innovation

Québec’s AI strategy – and the need to bring the workforce with you

Talk turned to the importance of garnering workforce support in the deployment of responsible AI.

Sébastien Rivard is director of the Artificial Intelligence Centre of Expertise at the Government of Québec’s Ministry of Cybersecurity and Digital Affairs.

“I’ll speak for the perspective of the Government of Québec, but as you know, we all share the common goal in Canada,” he said, before describing the AI strategy Québec has developed and is in the process of applying.

“We [have been] building, understanding guidelines, guardrails, best practice for years, focusing on increasing our capacity to deliver to the citizen through the usage of AI, [with a focus on] the responsible use of AI,” he said.

“We didn’t go all-in, we put a lot of thought into it. So, AI is new in terms of [the government’s] usage of it, but what isn’t new is the understanding of how AI can be used to [benefit our citizens].” 

Rivard explained that training in AI is a core component of building trust among the workforce, and aims not just to make staff more technically proficient but also more ethically sensitive. Workforce empowerment related to AI comes from making sure there is always a human in the loop, and that humans really are “in charge” and “responsible for the process”, he said.

The government need its employees to embrace the adoption of AI, and for them to do this “we need them to understand the rules, to feel confident about it, and to deliver the best service possible to the citizen”, he added.

Christine Jackson is executive director for the Canadian public sector at SAS. She said she had enjoyed the privilege of seeing “how AI adoption and rollout is occurring at all levels of government” in Canada, and that what she’d noticed is that AI adoption differs by province, at the federal level, and “even inside departments within each of those facets of government”.

Following on from Rivard’s point about the workforce, she emphasised that: “One of the most important things that government can remember is that AI transformation is as much about the technology transformation as it is the people transformation”.  

Read more: Canada publishes new national AI strategy with focus on trust and sovereignty

Emphasis on human-enabled AI

Successful implementation of AI in government depends on whether employees trust in, understand, and feel they’re part of their employer’s AI journey – and on allaying fears that they may be at risk of being replaced by it.  

“The way I like to think of this is that it’s an ‘and’ strategy versus an ‘or’ strategy,” Jackson said. “The emphasis needs to be on humans enabled by AI, as opposed to humans versus AI. Ultimately, employees and citizens are going to be much more open to AI tools when they see them reducing some of the monotonous administrative pieces [of work and life].”

She added a point about explainability. “If a public servant can’t explain why an AI-assisted recommendation was made, or how the data was informed, chances are that that process probably shouldn’t be rolled out for broad deployment yet.”

“We talk a lot with our customers about explainability and governance-by-design, as opposed to it being an afterthought, and… when that’s at the forefront… it becomes a lot more beneficial,” she said, though she noted that there is “still work to be done”.

Jackson has seen effective examples of explainability and governance-by-design in what she called the “low hanging fruit”: low risk, high value use cases that not only accomplish important small tasks, such as summarising documents, but also achieve an overall increase in trust and support for the government’s mandate.

This, she said, is where she and her colleagues are seeing real buy-in for AI in government.

Read more: Overcoming barriers to AI to unleash transformation in government

More meaningful work

Maher Mamhikoff, director of data, artificial intelligence and performance at Global Affairs Canada, agreed with what Jackson had said about buy-in being achieved when public servants can see that AI is making tasks easier, and their jobs more interesting.

He gave examples of the kinds of more meaningful work public servants would be freer to engage in if more government processes were successfully automated.

Analysts at Global Affairs Canada tend to “wear many different hats”, not only advising and working with local communities to review the results of the department’s projects but also engaging in policy development.

He gave the example of gender-based analysis of foreign assistance projects, which can take up a lot of analysts’ time. 

“What we’re looking to do is enable AI to do that [gender-based analysis] much quicker. When an expert is looking at a 100-page document, the AI can quickly tease out the most prevalent pieces, allowing the analyst to zoom in and make their work more concrete, and to complete it faster,” he said.

Automating this process frees the analyst to discuss the results of the project with their partners on the ground in greater depth – dialogues that are key to building trust and to understanding how the organisation can improve its approach to integrating gender equality into its projects, Mamhikoff explained.

The time saved on extracting relevant information from documents also allows more time for scrutinising the evidence gathered and for policy development, granting experts at Global Affairs Canada greater work satisfaction.

Throughout the conversation, panellists agreed that the responsible deployment of AI was about realising all the ways in which AI could serve all people, from the government official to the citizen accessing government services.

Read more: New Zealand government forges path to responsible AI with new framework

Striking the right balance – and ‘strong and nimble’ AI governance

The Government of Canada and many of the country’s provincial governments have been robust in their pursuit of rolling out responsible AI.

As Mamhikoff said: “We’re hitting that check and balance within the government on the security, the platform, all these elements that come into play before you’re able to deploy.”

And this, he added, presented a challenge: that by implementing so many checks and balances, government risked creating a pipeline of AI that takes so long to deploy that it inhibits it from realising its goals.

To reach the right level of AI maturity, Mamhikoff said, government would need to mitigate this risk.

But he said: “You have to build the governance. Governance not only of the intake, but what it does, the management of these multitude of AI solutions that you will have running. What does that management look like for the next number of years? And how do you keep those human elements, the experts, the technicians, there to make sure that it continues to be accurate, and useful, and provides value for Canadians?”

Dykes offered her thoughts on what she described as the importance of “strong and nimble” AI governance.

“We’re talking about government-wide governance around AI tools, but also within departments, so that you have not just the right tools being developed, but you also have the right checks and balances, and people always in the loop.”

She concluded: “It’s good in a way that [the private sector] can take some of the risks, but we do have to catch up. We can’t have the type of tech debt that we have had in the past. That won’t be sustainable.”

The ‘Responsible AI in government: What’s working in Canada’ webinar was hosted by Global Government Forum, with support from knowledge partner SAS, on 21 May 2026. Watch the webinar in full here.