Canada launches consultation on AI transparency

The federal government has launched a consultation on AI Transparency as part of the trust-building pillar of its new AI Strategy. The stated objective of the transparency initiative is to help Canadians “better understand when and how they are interacting with AI systems and AI-generated content”. Responses are due by September 23, 2026.

The consultation document addresses five potential action areas for the government. These are:

· Detecting and identifying AI-generated content;

· Empowering individuals to know when they are interacting with an AI system;

· Improving the availability of consistent and understandable information about AI systems, including their development, capabilities, and limitations;

· Enabling the tracking of serious incidents related to AI systems; and

· Advancing ways to better track the activity and interactions of AI agents.

This consultation is welcome. The document acknowledges both the newness of the area and its rapid evolution – these features make the consultation particularly important as there may be much that the government does not yet know about how Canadians are interacting with AI and how they are experiencing those interactions. Beyond this, consultation is an important part of advancing AI technical and policy literacy. The document highlights some of the features of AI systems that raise concerns: how rapidly they evolve and how difficult it may be to pinpoint where regulation is needed and where it would be counterproductive. It also explains some of the challenges faced by the government in addressing these issues appropriately, and it discusses a much broader range of policy options than traditional legislation. It sets out some of what the government is already working on, and touches on what some other jurisdictions are doing on specific issues, noting that it is still too early to assess the successes or problems with these approaches. It also discusses some companies’ efforts to improve transparency in different contexts. Overall, the document is a useful, thoughtful piece that seeks input – not on specific planned measures – but on overall needs, strategies, approaches, and concerns.

It is noteworthy that the document continues with the current trend internationally that has shifted the conversation from ‘responsible AI’ to ‘AI safety’. The focus is thus on bigger, more visible harms. More complex and less visible risks may receive less attention. For example, using the AI safety lens, the document mentions risks of system malfunctions that can “result in biased decisions.” However, biased outcomes may not be the result of a malfunction; rather, they could be the result of an uncritical use of unsuitable data, biased assumptions in the design of the algorithm, or even problematic choices as to where and how to deploy AI systems. Further, although the document notes that “researchers and auditors may need information about training data to assess a system for bias or other risks”, it is interesting to note that the emphasis here is on transparency as to system inputs and not outputs. This highlights a major challenge when it comes to AI transparency – how do we prioritize what we need to know about these systems, and who needs to know what, how, and for what purposes?

The consultation document rightly states that transparency can mean different things in different contexts for different actors. Workers, consumers, regulators, and researchers may all need access to different types of information. The document notes that “Transparency will not address all risks from AI, but it can be a foundation for informed decision-making, accountability, good business practices, and – where appropriate – further government intervention”. All of this is true. Yet, the “further government intervention” piece is a reminder that transparency requirements are only a first step. For example, the document mentions the new obligation on employers of a certain size to give notice of the use of an AI system in a publicly advertised hiring process in Ontario’s Working for Workers Four Act. However, measures such as this offer relatively little beyond that notification. Bare transparency can let you know that an AI system is being used to determine whether you get an apartment, a job, or a particular price, but unless there is some capacity to assess whether the system is producing biased or exploitative outcomes, the broader human rights and public policy issues are left unaddressed. The point is that while individual-level transparency is desirable, systemic transparency is also important for certain issues and may require different measures and greater oversight.

Clearly, it is all a work in progress, and it is daunting to think of just how much work there is to do in this space. That said, the consultation is important, and the consultation document casts its net widely in an attempt to gather a broad range of feedback and input. These are good things. The government is clearly trying to chart a path at a time of great economic instability and in a technological environment that changes rapidly and that frankly is largely controlled by forces outside Canada. It is not an easy task. What it means, though is that as the tech giants cling to their “move fast and break things approach”, the government’s response is to “move slow and try not to break anything”. The AI safety lens is meant to focus regulatory attention on the biggest things that can do the most harm. Nevertheless, the potential for harm in this context is significant and not completely knowable. It is also very relative. The harms experienced by the most vulnerable, and the harms that are least visible and more complex are often the ones that are neglected until they manifest in shocking ways – as we are learning now with our very late response to the harms of social media on children and youth.

This broad, information gathering consultation on AI transparency is an opportunity to flag issues and raise concerns. Be sure to have your say by September 23, 2026.