Process as Assurance: AI Governance in the UK and Canada

Process as Assurance: AI Governance in the UK and Canada

Co-authored by Carole Piovesan, Managing Partner, INQ Law (Toronto, Canada).

UK-Canada AI Security Partnership

On 13 October 2026, Canada and the UK renewed their AI security partnership, first signed in 2024 as the AI Safety Partnership, with the aim of fostering the safe and responsible development and deployment of AI.

The renewed agreement commits both countries to focus on three priorities:

  1. Interoperable standards: developing mutually recognised or shared AI security standards; 
  2. Enhanced information-sharing: improved information exchange, including on frontier risk; and 
  3. Closer technical cooperation: expanding technical arrangements, including regular exchanges and coordinated evaluation activities between the Canadian AI Safety Institute and the UK AI Security Institute.

Since its inception, Canada and the UK have worked closely on various AI initiatives, including the UK AI Security Institute’s Alignment Project, to which the Canadian Institute of Advanced Research contributed CAD $1 million. 

Why This Matters

While the UK and Canada are working more closely on AI security, neither country has passed a single, comprehensive law that governs AI across every sector. Canada came closest with a proposed Artificial Intelligence and Data Act which was abandoned when Parliament was dissolved in 2025. The UK has deliberately chosen not to legislate. Instead, organisations must navigate a patchwork of compliance obligations across existing laws, regulations and standards, and be able to demonstrate diligence in the use of higher-risk AI use cases. 

No legislation in the UK or Canada sets out exactly how organisations must assess the risks associated with developing and deploying AI, but both regimes are asking the same questions: has the impact of the AI system been assessed before deployment? Have decisions been documented? Is there meaningful human oversight? Will there be ongoing monitoring? 

Our recommendation is that implementing a consistent and repeatable process to answer these questions will allow organisations to respond to the AI security partnership priorities, evidence compliance with applicable regulatory regimes in both jurisdictions, and be prepared for any future legislation.

Two Systems, One Pattern

Canadian Position

A patchwork, not a single statute

Canada’s legal environment for AI is fragmented across jurisdictions and sectors. Canada does not have a comprehensive, cross-sector law specific to AI. Instead, organisations must navigate a patchwork of guidance, sectoral regulation and laws of general application - including privacy, human rights, consumer protection and competition law - each of which has a bearing on how AI is designed, procured and deployed.

Several soft-law instruments set expectations in the meantime:

  • Voluntary Code of Conduct: the federal Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems sets out commitments on accountability, safety, fairness, transparency, human oversight and robustness;

  • Directive on Automated Decision-Making: binding on federal institutions, the Directive is supported by the Algorithmic Impact Assessment (“AIA”), a tool that scores a system’s risk and scales the required safeguards - peer review, notice, human involvement and explanation - accordingly.

Draft law on the table

In June 2026, the federal government tabled Bill C-36, which would enact the Protecting Privacy and Consumer Data Act as Canada’s new private-sector privacy law. Among other things, it would require organisations to disclose their use of automated decision systems that could have a legal or similarly significant effect on individuals, and allow individuals to obtain an explanation of, and challenge, those decisions. A companion bill, the Safe Social Media Act (Bill C-34), would impose digital safety duties on social media platforms and chatbot services. Both regimes would be overseen by a new Digital Safety and Data Protection Commission of Canada.

Sectoral regulators are moving

  • Health Canada: has issued pre-market guidance for machine learning-enabled medical devices, setting expectations for AI-based software as a medical device across the product lifecycle;

  • Office of the Superintendent of Financial Institutions (OSFI): Guideline E-23 on Model Risk Management, in force from 1 May 2027, expressly captures AI and machine learning models and expects governance that scales with model risk at federally regulated financial institutions;

  • Standards Council of Canada: has convened work on AI standards and trust frameworks, laying the groundwork for conformity assessment and certification.

From research to adoption

In 2017, Canada became the first country to launch a national AI strategy, focused on research, talent and responsible AI. Its successor, AI for All, launched in June 2026, shifts the emphasis to adoption: it aims to lift business AI adoption from just over 12% to 60% by 2034, while building trust through modernised privacy and online safety laws and an expanded Canadian AI Safety Institute.

Canada also now has a dedicated Minister of Artificial Intelligence and Digital Innovation. The Minister’s priorities flow from the Prime Minister’s May 2025 mandate letter to Cabinet, which calls on government to become “much more productive by deploying AI at scale” and ties the transformative nature of AI to new career opportunities for Canadians. The Minister has since introduced Bill C-36 (see above).

Budget 2026, tabled on 5 November 2026, backs this agenda with new commitments, including access to 500 MW of sovereign AI compute; approximately CAD $1 billion over five years for a Government Transformation Fund, alongside a new Digital Transformation Canada to lead AI adoption across the public service; CAD $176.3 million over five years to establish the Digital Safety and Data Protection Commission; and a plan to explore a Technology Stability Board to assess whether advanced AI models are safe and remain under human control.

What this means for assurance

No single Canadian instrument tells organisations how to assure their AI. Read together, however, the same core expectations are mirrored: assess impact before deployment, document decisions, keep humans in the loop, explain outcomes and monitor models across their lifecycle. The AIA and OSFI’s E-23 already function as practical assurance templates. Organisations that adopt one consistent, repeatable process aligned to them will be able to evidence diligence to regulators across sectors today, and be ready as Bill C-36 and the wider framework take shape.

UK Position

A principles-based approach

With no single dedicated legislation that specifically regulates AI, the UK has adopted a broad principles-based approach (namely: safety, security and robustness; appropriate transparency and explainability; fairness; accountability and governance; and contestability and redress), enforced by sector regulators. Many regulators have published their own sector specific guidance on AI, including:

  • Financial Conduct Authority (FCA): its AI Update outlines how its existing financial services rules apply to AI;

  • The Information Commission: the UK’s data protection regulator has issued guidance on AI and data protection and explaining decisions made with AI and has been tasked with preparing a formal Code of Practice on AI and automated decision-making; 

  • Competition Markets Authority: its AI Strategic Update sets out six principles aimed at fostering an environment conducive to competition and consumer protection. 

Data protection impact assessments

The importance of transparency and accountability runs across this regulatory guidance - key principles that also underpin UK Data Protection Law. If an AI system uses personal data, a data protection impact assessment ("DPIA") is one of the most useful tools available to assess risk, and evidence demonstrable assurance that these principles are being upheld. 

DPIAs provide organisations with a valuable opportunity to consider and demonstrate accountability for decisions made in the design or procurement of AI systems where this may involve the processing of personal data. Rather than treating DPIAs as a box-ticking compliance exercise, organisations should be utilising them as practical tools to identify and manage the risks to rights that may arise from the design, deployment or use of AI. They can therefore serve as a useful means by which organisations can evidence to regulators that they have carefully considered the impact of deploying an AI system.

Automated Decision-Making 

An interesting area of application, and one where the UK has recently diverged from the EU, is the UK’s approach to solely automated decision-making (“ADM”). Reforms under the Data (Use and Access) Act 2025 removed the general prohibition on ADM that produces legal or similarly significant effects, replacing it with a requirement for a lawful basis, such as legitimate interests, provided that no special category data is involved.

This change is particularly relevant for sectors that rely on algorithmic decision-making, including credit, fraud detection, recruitment, and pricing. However, organisations must still implement safeguards: ensuring transparency, providing individuals with the opportunity to make representations, allowing access to human intervention, and enabling individuals to challenge decisions. These are, in substance, the same safeguards that Canada’s AIA scales by risk and that Bill C-36 would extend to the private sector.

European Union: the EU AI Act

It is important not to overlook the impact of other applicable regimes. The EU, unlike Canada and the UK, has chosen to legislate specifically for AI. The EU AI Act’s broad extraterritorial reach means that it applies to organisations wherever they are based, including in the UK and Canada, that place on the market or put into service AI systems or general-purpose AI ("GPAI") models in the EU, or that develop or use AI systems, where their outputs are used in the EU. 

The EU AI Act came into force on 1 August 2024. Its obligations have taken effect in stages, with some provisions already in effect and others subject to deferred timelines as a result of the EU’s ‘digital omnibus on AI’, which has extended several key deadlines for AI-related obligations under the EU AI Act. This includes the application of rules for high-risk AI systems, originally due to take effect from 2 August 2026 but now deferred until 2 December 2027 (Annex III) and 2 August 2028 (Annex I). 

Assurance built into the law

It is these rules that offer the clearest demonstration of assurance built directly into the law. Providers of high-risk AI systems must, among other requirements, operate a robust risk management system; keep detailed technical documentation and records; design systems to allow effective human oversight; demonstrate compliance via a conformity assessment before the system is placed on the market; and conduct most-market monitoring. 

Deployers also have defined obligations, including assigning human oversight to trained staff, a statutory requirement to conduct a DPIA where required and, for public bodies and certain public services, carrying out a fundamental rights impact assessment. On ADM specifically, the EU retains the GDPR’s general prohibition on solely automated decisions with legal or similarly significant effects (subject to limited exceptions), and the EU AI Act adds a right for affected individuals to obtain an explanation of decisions based on the output of certain high-risk AI systems.

The reach of the EU AI Act means that many non-EU organisations will need to comply with EU requirements, and meeting the EU standard is likely to remain the benchmark, notwithstanding the UK’s recent relaxation of the rules on ADM. While compliance with the more prescriptive requirements of the EU AI Act will not automatically mean that an organisation complies with existing UK and Canadian rules, meeting the EU AI Act’s standards is likely to satisfy most local requirements and align with applicable UK and Canadian regulation and guidance.

Final Thoughts

Despite no instrument in the UK and Canada telling organisations how to assure their AI, the regulators’ expectations across all three regimes clearly converge on the same points: assess impact, document decisions, provide human intervention, explain outcomes and allow challenge. The DPIA is the UK’s ready-made assurance template, performing much the same function as Canada’s AIA. The ICO’s upcoming statutory code on AI and ADM is likely to further formalise expectations for AI deployment.

That is why a repeatable process supports assurance. One assessment methodology applied consistently to each higher-risk AI use case gives organisations three things:

  • Evidence that can be presented to a regulator; 

  • Consistency, streamlining decision-making and making gaps easier to spot; and

  • Resilience: it is easier to update or adjust a process than build a new one;

and in a rapidly evolving landscape without a single rulebook, it is the process itself that can provide the assurance.

This article was prepared in collaboration with Carole Piovesan, Managing Partner, INQ Law (Toronto, Canada).

‘

The collaboration combines Canada’s AI research strengths with the UK’s expertise in AI testing and evaluation, supporting both countries’ efforts to understand and respond to emerging risks.’

Share Article

Related Expertise