Canada launches a public consultation on AI transparency, signaling a future of more frequent notice and more disclosure about risks
Last week, the federal government launched a consultation that will run until mid-September, inviting Canadians to have their say on transparency in artificial intelligence across a range of areas.
The department overseeing this — Innovation, Science and Economic Development Canada (ISED) — also published a detailed discussion paper , titled “Enhancing Trust in Artificial Intelligence Through Increased Transparency.”
Although I suspect it was produced with an AI tool like Deep Research , I found it well worth reading, for the quality of the overview on two fronts: the technological capabilities for transparency across AI and the existing law and policy on point in Canada, Europe, the US, and elsewhere.
At the moment, Canada has many laws touching on the risks and harms arising from AI, but we have no law on AI transparency specifically. Instead we have “The Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems ,” which is narrow in scope (only model developers) and voluntary.
The purpose of the present consultation is, as the discussion paper puts it, for the government to gain a better sense of “where transparency about AI systems and AI-generated content matters most to Canadians, where existing practices and frameworks are sufficient, and where further action may be warranted.”
The paper looks at five areas specifically: our encounters with AI-generated content; our interaction with AI tools; information about the models and systems themselves; incidents arising from AI tools; and AI agents.
The paper floats ideas about how to improve transparency on all these fronts. The ideas offer a glimpse of a possible future of AI in Canada in which model creators and users must give more disclosure about risks and more frequent notice about use.
This raises questions about what difference that might make in the way people produce and share content with AI or use it in their products and services, or the way companies create and deploy AI models and agents. Will they use it more sparingly, more responsibly, or more confidently?
Given the broad scope of ISED’s interest in transparency — as gleaned from the paper — it seems likely the consultation will lead to a new bill on AI focusing on transparency. And this could impact our experience of AI in meaningful ways, for better or for worse. But that’s getting ahead of the story.
What we can glean from the paper are the key areas of interest to the government, noted above, along with possible approaches it might take to increasing transparency.
The paper’s general structure is to provide a snapshot of the state of the technology for boosting transparency (or impediments to this) in each area, existing law on point here and elsewhere, and a segment that floats possibilities for law or policy the federal government might adopt. And here, the paper poses many thoughtful questions.
To highlight a few points that stand out, the segment on mandating greater transparency around AI-generated content notes the difficulty of defining what counts as ‘AI-generated.’ How much reliance on AI in producing an image, text, or piece of music makes it AI-generated? When a journalist has AI produce a draft of an article but substantially revises it, is it still ‘AI-generated’?
This very issue has come to the forefront this week here on Substack itself, following its new partnership with Pangram to offer an AI detection tool in the Substack app. A number of authors have questioned how well this works, with content flagged as ‘100% AI generated’ when it was only copyedited with AI, or having it labelled ‘100% human’ when whole paragraphs are the output of AI. Defining what is ‘AI generated’ is therefore both a technical and theoretical problem we haven’t solved, but a pressing one with much at stake for those having their work mislabelled.
The discussion paper points out that the Safe Social Media Act, in Bill C-34 , will require regulated social media platforms to label synthetic content “likely to be mistaken for an authentic visual or audio recording of a person, object, place, entity, or event.” But this applies only on platforms where content is shared. It doesn’t apply to text, and the paper flags that it “does not create obligations on developers and deployers of AI systems used to generate this content.”
Could this be one direction a new law might go? Not a disclosure obligation for those producing AI-generated text, but maybe one for systems or deployers of AI generated audio or visual material, on the test of whether content is “likely to be mistaken” for an authentic recording of a person or event, possibly facilitated through a watermark or metadata?
When it comes to interacting with an AI system, the paper’s main concerns are with using AI for customer service and for assessing job applications. In the latter case, the paper hints at the prospect of adding disclosure obligations in federal legislation mirroring those in Ontario’s employment law. For customer service AI, the paper is less clear about how to proceed, posing questions about when disclosure is appropriate, and how much.
For the makers and deployers of AI systems — models and chatbots — the paper is concerned with companies being transparent about what “an AI system can and cannot do (its capabilities, limitations, and appropriate uses)” and about “how it was designed, including the data on which it was trained.” One issue arising here is how to balance adequate disclosure and the protection of trade secrets. Another is whether more disclosure about model limitations (I am not a doctor or lawyer) will help us make more informed choices about how we use AI.
The paper also highlights a need for more transparency around AI incidents involving harm, failure, or fraud — an issue in the headlines last week, with news of an OpenAI model breaking out of a testing sandbox and hacking into Hugging Face’s servers with a stolen password. The paper notes that while Canadian law already contains reporting requirements in several domains (health, transport, consumer products, critical cyber systems), we lack a framework for ‘AI incidents.’ We could benefit from a statutory definition and threshold for serious incidents triggering disclosure obligations.
Finally, the paper canvasses the prospect of increasing transparency around the use of AI agents. The main concerns here are that “users may not always know what permissions have been granted to the agent, when an AI system is acting on their behalf, whether a human remains in control, or how to challenge or reverse an action that an agent has taken.” Tools can malfunction or be misused or hacked. At the moment, consumer protection, contract, and tort law would form the basis of liability for harms arising here. But greater disclosure of the risks and limitations of these agents could lead to safer and more responsible use and deployment of them.
The government last conducted a major public consultation on AI in October 2025, for its national AI strategy. That garnered some 11,300 responses from businesses, institutions, NGOs, and individuals, resulting in Canada’s “AI for All ” strategy unveiled in June 2026.
I suspect the transparency consultation won’t generate the same number of responses, but it might result in a report as informative as “AI for All,” and possibly a bill that aims to improve transparency in one or more of the areas canvassed here. ■
{To receive new posts by email, follow on Substack .}