Canada's AI and Copyright Consultation: the Outcome So Far

Canada consulted on generative AI and copyright, published what it heard, and has not legislated. That is the outcome, and stating it plainly is more useful than the alternative, because the practical question — what governs training on copyrighted works in Canada today — has an answer that has not moved since 2023.

What the consultation asked

Innovation, Science and Economic Development Canada, together with the Department of Canadian Heritage, launched a consultation on copyright and generative artificial intelligence on 12 October 2023, with submissions closing on 15 January 2024. The consultation paper organised the ground into three questions, and the framing is worth knowing because it maps onto the fault lines everywhere else:

• Text and data mining. Whether Canada should adopt an exception permitting reproduction for machine analysis, and if so whether it should be limited to non-commercial use, conditioned on lawful access, or subject to an opt-out.
• Authorship and ownership of AI-generated works. Whether output with no human author should be protected, left unprotected, or given a limited sui generis term.
• Infringement and liability. Who is responsible when an AI system produces infringing output — developer, deployer or user — and whether existing remedies are adequate.

The consultation material and subsequent copyright policy documents are published by ISED’s copyright policy directorate.

What has come of it

The government received a large volume of submissions, published a summary of what it heard, and has not brought forward an amendment to the Copyright Act on any of the three questions. As at the date on this page, Canada has no text-and-data-mining exception, no statutory rule on authorship of AI-generated works, and no AI-specific liability provision in its copyright statute.

Two contextual facts explain some of the delay without excusing a prediction. The consultation ran across a period in which the government’s other major AI legislative vehicle collapsed, and Canadian copyright reform has a long history of moving in large multi-year packages rather than in single-issue amendments — the last substantial modernisation before this was the Copyright Modernization Act of 2012, and the statutory review of the Act reported in 2019 without leading to a general amending bill. It is not yet clear whether the AI questions will be legislated separately or folded into a future package, and nobody outside the process can say which.

The law that therefore still applies

In the absence of amendment, training on copyrighted works in Canada is governed by the Copyright Act as it stands. The relevant features:

Fair dealing under sections 29, 29.1 and 29.2 is available for research, private study, education, parody, satire, criticism, review and news reporting. Unlike Australia’s, Canada’s list has been read generously: the Supreme Court of Canada held in CCH Canadian Ltd v Law Society of Upper Canada in 2004 that the allowable purposes must be given a large and liberal interpretation and that research is not confined to non-commercial contexts, and it has repeated that framing since. That makes a Canadian fair dealing argument for training more plausible than an Australian one, and considerably less certain than an American fair-use one, because a court would still have to accept that commercial model training is “research” and then run the six-factor fairness analysis from CCH.

The honest statement is that this is untested. No Canadian court has decided whether training a generative model is fair dealing. Canadian news organisations have filed proceedings against OpenAI in Ontario, and the early skirmishing in that litigation has been about jurisdiction rather than the merits, so it has produced no ruling on the copyright question. Anyone telling you the Canadian answer with confidence is telling you a view, not a holding.

Why AIDA is a separate story

The Artificial Intelligence and Data Act was Part 3 of Bill C-27 and had nothing to do with copyright — it was a risk-based regulatory framework for high-impact AI systems, in the general family of the EU AI Act. It died when Parliament was prorogued in January 2025, which means it would have to be reintroduced from scratch to proceed.

The two tracks are frequently merged in commentary, with the result that people conclude Canada “has no AI law” and stop looking. That is wrong in the direction that matters: federal privacy law under PIPEDA and provincial regimes including Quebec’s Law 25 apply to AI systems processing personal information today, with no new statute required. See the status of AIDA and the Privacy Commissioner’s guidance.

What would change the answer

Three things, in rough order of likelihood, and each is checkable rather than speculative:

• A bill. An amendment to the Copyright Act introducing a TDM exception or an authorship rule. Watch the Parliament of Canada LEGISinfo tracker; a bill’s text at first reading tells you the shape long before it passes, and most do not pass.
• A judgment. A ruling in the Ontario proceedings against OpenAI on fair dealing would be the first Canadian authority on the point, and would be appealable to the Court of Appeal for Ontario and potentially to the Supreme Court.
• An international commitment. Canadian copyright policy has historically moved in response to trade obligations, and a provision on AI and copyright in a future agreement would drive domestic amendment faster than a consultation.

For the comparable UK process, which asked similar questions on a different timetable, see the UK AI and copyright consultation outcome.