Chapter 3 established that an AI model is, in a meaningful practical sense, its training data, rearranged into compressed statistical form. Chapter 9 introduced personal sovereignty as a distinct layer of the sovereignty conversation, with consent for AI training as one of its three central domains. Chapter 10 mapped the broader privacy and surveillance landscape and the post-AIDA regulatory gap. This chapter takes those threads and goes deep on the specific Canadian intellectual property contest. The contest is not abstract. As of mid-2026 there is an active Ontario Superior Court case, Toronto Star Newspapers Ltd. et al. v. OpenAI, Inc., concerning approximately 16.1 million Canadian news articles that the plaintiffs allege were used to train OpenAI's models without licensing. There are formal positions from the Writers' Union of Canada and ACTRA opposing AI training without consent. There are detailed academic arguments from Carys Craig at Osgoode Hall and Michael Geist at Ottawa arguing that AI training should be permitted under a text-and-data-mining exception to Canadian copyright. There is the documented observation — TWUC's primary documents make it directly — that Canadian copyright has been "severely weakened through Supreme Court of Canada decisions over the past 15 years and government inaction to repair that damage." And there is the federal AI strategy's structural silence on creator IP rights. This chapter does what the guide's methodology requires: presents the genuine Canadian academic and legal disagreement honestly, without flattening any side into a strawman. Geist and Craig are rigorous, principled scholars whose position is grounded in serious analysis of how copyright law actually works. TWUC and ACTRA are rigorous, principled advocacy organizations whose position is grounded in serious analysis of how creator economies actually work. They reach different conclusions because they prioritize different values — and the disagreement is the chapter's central teaching example for how Canadian AI policy debates should be read. You will leave with: the Canadian newspapers v. OpenAI case structurally understood; the Geist/Craig vs TWUC/ACTRA academic-and-advocacy disagreement engaged in both directions; the structural finding that Canadian copyright is itself in a contested state predating AI; the NIL Rights / Beijing Treaty argument as the specific Canadian creator-side proposal; the working test for evaluating any AI copyright claim; and the conceptual handoff to Chapter 20, where the path-forward implications of all these positions get specific. ---
The Canadian newspapers v. OpenAI case
Filed November 29, 2024 in the Ontario Superior Court of Justice, case number CV-24-00732231-00CL. Plaintiffs: CBC/Radio-Canada, the Canadian Press, Torstar, Postmedia, Metroland Media Group, and the Globe and Mail. Defendants: OpenAI, Inc. and OpenAI Canada, ULC. Approximately 16.1 million owned and licensed Canadian news works are at issue — by article count, the largest copyright claim in Canadian history.
The plaintiffs' core claim, in plain terms: OpenAI used 16.1 million Canadian news articles to train its commercial AI products, without licensing the use, without compensating the rights holders, and without engaging the collective licensing infrastructure that exists in Canada for similar uses of copyrighted news content. The plaintiffs allege copyright infringement under sections 3, 27, and other provisions of the Copyright Act of Canada, and seek damages, punitive damages, payment of profits attributable to the infringement, and an injunction against further use of the plaintiffs' works.
OpenAI's defence, in compressed form: the training of AI models on publicly available content is a permissible use under existing copyright frameworks. The company has argued that its training practices fall within fair-use-style exceptions, that the resulting models do not "reproduce" the training data in the conventional copyright sense, and that the policy question of how to govern AI training data is properly a matter for legislative action rather than judicial determination.
The key procedural finding. On November 7, 2025, the Ontario Superior Court rejected OpenAI's jurisdictional challenge. OpenAI has since appealed that ruling, so where the case is ultimately heard is not yet settled — but for now an Ontario court has claimed the jurisdiction OpenAI tried to deny it. This is significant because OpenAI had argued that the case should be heard in the United States, where most of OpenAI's operations occur and where similar litigation (The New York Times v. OpenAI, Authors Guild v. OpenAI, and others) is already underway. The Ontario court's decision to take jurisdiction makes Canada one of the first jurisdictions globally to formally adjudicate AI training copyright in a major commercial case. The substantive trial has not yet occurred as of this writing.
What the case will and won't decide. The case will decide whether OpenAI's training practices violated Canadian copyright with respect to the specific 16.1 million news works at issue. It will not decide, in any direct way:
- Whether AI training on copyrighted material is generally permissible under Canadian law (the case is about specific works and a specific defendant, though the reasoning will be precedential)
- Whether other Canadian creator categories (writers, artists, musicians, photographers) have valid claims (those would require separate cases)
- Whether the Canadian Copyright Act should be amended to address AI training (that is a legislative question)
- Whether OpenAI is broadly liable across all its training practices globally (that depends on jurisdictions and on specific corporate structures)
The case's significance is broader than its specific holding. It will be the first detailed Canadian judicial analysis of how Canadian copyright law applies to AI training, and its reasoning will be cited in every subsequent Canadian case. The reasoning may favour the plaintiffs, may favour OpenAI, or may produce a mixed result with damages awarded on some grounds and rejected on others. The reasoning will shape Canadian policy for years.
EXAMPLE — the cook who read ten thousand cookbooks. A chef reads every cookbook in the library and writes a new recipe. Did they copy, or did they learn? The authors say their words are baked into that dish; the chef says the dish is original. Canadian courts haven't yet decided which description wins when the "cook" is a model and the ten thousand cookbooks were copied without anyone asking.
The political-economy reading of why this case matters to Canadian news specifically. Canadian news organizations have been in a sustained financial decline driven by advertising revenue migration to Facebook and Google over the past two decades. The federal government has responded with the Online News Act (which prompted Meta to block Canadian news on Facebook and Instagram in 2023). News organizations are now facing AI training as a second front in the same broader contest: digital platforms extracting value from journalism while Canadian news economics deteriorate. The lawsuit is one of the few mechanisms by which Canadian news organizations can attempt to recoup value from the AI extraction. Whether they succeed in this case, and on what terms, will substantially shape whether Canadian news organizations have a viable path through the AI moment.
---
The TWUC and ACTRA positions
Beyond the active litigation, two Canadian creator organizations have published formal positions on AI training and copyright that constitute the most-detailed Canadian creator-side documentation: the Writers' Union of Canada (TWUC) and the Alliance of Canadian Cinema, Television and Radio Artists (ACTRA).
The Writers' Union of Canada position.
TWUC's formal position, available at writersunion.ca/advocacy/artificial-intelligence and in their January 2024 submission to the ISED Copyright in the Age of Generative AI consultation, can be summarized in two structural claims:
The first claim is about inputs. AI training, including text-and-data mining, on published material without author permission constitutes copyright infringement under existing Canadian law. The claim is grounded in the analysis that training is a "reproduction" (section 3 of the Copyright Act) of the training data, that the reproduction is for a commercial purpose, that no statutory exception clearly applies, and that the existing fair-dealing exceptions (section 29) require a much narrower purpose-specific use than AI training represents.
The second is about outputs. AI outputs that do not involve significant human creativity should not be eligible for copyright protection. The claim is grounded in the analysis that copyright is designed to incentivize human creative work, that machine-generated outputs without substantial human direction don't engage that rationale, and that allowing AI-generated content to receive copyright protection would functionally extend AI training's value extraction to also extract from the public domain.
TWUC's specific actions on the ground:
- Revised Model Trade Book Contract with an explicit author opt-out clause for AI training rights. This is the standard contract TWUC recommends authors use with publishers. Canadian authors using the TWUC model contract have, since 2024, had a standard mechanism for refusing AI training rights distinct from authors in many other jurisdictions.
- Two Canadian class-action lawsuits being prepared as of TWUC's most recent public communication. "Any Canadian author with work in the contested datasets would almost certainly be part of those classes."
- May 2024 coalition of 10 writing, publishing, and creator organizations, spanning English and French Canada and representing 300+ Canadian book publishers and 10,000+ creators, met Ottawa parliamentarians around a book display of Canadian titles used without consent in AI training. This is the most significant Canadian creator-coalition primary event on AI in the period covered by this guide.
- Cautious response to the HarperCollins Canada AI licensing proposal. TWUC Chair Danny Ramadan has been publicly "cautiously optimistic" about negotiated licensing as one pathway, distinct from rejecting all AI use of copyrighted material. The position objects not to licensing but to uncompensated extraction.
A specific TWUC framing the guide notes as primary-source: TWUC's January 2025 industry communications include the observation that "Canadian copyright law has been severely weakened through Supreme Court of Canada decisions over the past 15 years and government inaction to repair that damage." This is a significant claim — that Canadian copyright is already in a degraded state before AI even arrived. The Supreme Court decisions in question include the 2012 "copyright pentalogy" cases (CCH Canadian v. Law Society of Upper Canada, Society of Composers, Authors and Music Publishers of Canada v. Bell Canada, Alberta (Education) v. Canadian Copyright Licensing Agency, and others) that significantly expanded the fair-dealing exception in Canada. TWUC's argument is that the resulting expanded permissions plus federal inaction on creator-rights legislation have produced a Canadian copyright environment substantially less protective than peer jurisdictions. The federal AI strategy's silence on creator IP rights compounds the structural weakening.
The ACTRA position.
ACTRA's position, documented in their 2024 AI Submission Explainer and in subsequent collective bargaining outcomes, focuses on the specific creator category ACTRA represents: 28,000+ Canadian performers across film, television, radio, and digital media. The empirical baseline from their 2023 internal survey: 98% of surveyed members reported concern about potential misuse of their name, image, and likeness rights; 93% expected AI to eventually replace human actors in certain roles. (The survey went to all 28,000+ members; ACTRA did not disclose the response count — a scope note the guide's own methodology requires.)
ACTRA's position is structurally organized around the NIL Rights framework: the formalized legal claim to Name, Image, and Likeness rights borrowed from US sports and entertainment law and increasingly used in international creator-rights advocacy. The core claim: a performer's name, image, voice, and likeness are economic assets that the performer should retain rights over, including rights against unauthorized AI-generated reproductions.
ACTRA's specific operational actions:
- Independent Production Agreement (IPA) ratified January 21, 2025 — the first Canadian collective agreement with explicit AI provisions for performers. The agreement prohibits use of a performer's recordings to "simulate or alter a Performer's voice or likeness; to create any synthesized performance or 'digital double' voice or likeness of a Performer; or for machine learning." This is binding contract language in Canadian performer collective bargaining, ratified after the Hollywood SAG-AFTRA 118-day strike (2023) established the international precedent.
- Joint testimony with the Directors Guild of Canada (DGC) and Music Canada to the Bill C-27 parliamentary committee on February 12, 2024 (INDU meeting 110). DGC National Executive Director Dave Forget proposed three specific requirements drawn from the EU AI Act: (1) authorization of rights holders for the use of copyrighted content in AI training; (2) transparency requirements for general-purpose AI systems including disclosure of training materials; (3) compliance with Canadian copyright law including consent for data mining.
- Beijing Treaty implementation argument. ACTRA has argued that Canada's failure to implement the Beijing Treaty on Audiovisual Performances (a 2012 WIPO treaty) leaves Canadian audiovisual performers with weaker moral rights than sound-recording performers. The Beijing Treaty would extend moral rights (the right to be identified as the performer and the right to object to derogatory treatment of the performance) to audiovisual contexts including AI-generated reproductions. Canada has neither signed nor acceded to the treaty; it is not among the treaty's 48 contracting parties. This is a specific legislative gap with concrete remedy.
---
The Geist and Craig academic positions — and why they matter
Two Canadian academic voices, Michael Geist at the University of Ottawa and Carys Craig at Osgoode Hall, have published detailed analyses of Canadian copyright as it applies to AI training, reaching conclusions that differ substantially from the TWUC and ACTRA positions. This is not two camps talking past each other. Geist and Craig have both engaged the TWUC and ACTRA arguments directly and arrived at different conclusions because they prioritize different values within the copyright framework. The guide takes both positions seriously because both are rigorous.
Michael Geist's position, available in his January 2024 ISED consultation submission at michaelgeist.ca and in subsequent academic writing, can be summarized:
First: Canadian copyright already permits text-and-data mining (TDM) on copyrighted material for non-commercial research purposes through the fair-dealing exception (section 29) as expanded by the 2012 Supreme Court pentalogy. Geist's analysis is that the existing fair-dealing framework, applied to TDM specifically, supports broad permissibility of AI training, particularly for purposes of research and analysis.
Second: Even for commercial AI training, Canada should adopt an explicit TDM exception modelled on the European Union's TDM exception (Article 4 of the 2019 EU Digital Single Market Directive). The EU exception permits TDM on lawfully accessed material for any purpose, subject to a rights-holder opt-out mechanism. Geist's argument is that adopting this framework would (a) bring Canadian copyright into alignment with the international consensus emerging in the EU, UK, Japan, and Singapore; (b) avoid hobbling Canadian AI development by requiring licensing for activities permitted in competitor jurisdictions; (c) preserve rights-holder agency through the opt-out mechanism; and (d) ensure that Canadian AI development remains viable rather than being driven offshore.
GOBLIN FACTS — the exception Geist points to has an address. The EU model is Article 4 of the 2019 Copyright in the Digital Single Market Directive: it permits text-and-data mining of lawfully accessed work for any purpose, but lets rights-holders opt out and reserve their material. "Adopt the EU approach" means adopting that opt-out, not a blanket permission. Canada has neither.
Third: Geist's broader concern is with what he characterizes as the "copyright maximalism" of creator-organization positions, the argument that more copyright protection is always better for creators. Geist's analysis is that excessive copyright protection has historically benefited large rights aggregators (publishers, studios, record labels) more than individual creators, and that AI policy debates risk producing copyright extensions that primarily benefit corporate rights holders rather than individual writers, performers, or artists. The argument that "stronger copyright protects creators" requires examining who actually owns the copyright in question. In many cases, the answer is corporate publishers, not the individual creators whose names are used to make the political argument.
ALIGNMENT — licensed, or just unchallenged? When an AI company says its training is legal, notice which one it means: a court has agreed, a licence was signed, or no one has successfully sued yet. "Allowed" and "not yet stopped" look identical right up until a ruling lands. In Canada, most of this is still firmly the second one.
Carys Craig's position, available in her 2024 ISED submission (SSRN 4718941) and in her 2021 chapter in Florian Martin-Bariteau and Teresa Scassa's AI and Copyright edited collection, can be summarized:
Craig begins from the technical practice. AI training, on her analysis, does not constitute "reproduction" in the conventional copyright sense: training extracts patterns from a work and embeds them in a different kind of artifact (the model) rather than producing copies of the original, and treating it as reproduction misunderstands what training actually does — copyright law should respond to the practice rather than to a doctrinal category that doesn't fit. From there she argues that Canadian copyright should explicitly permit AI training through statutory exception, combining a broad fair-dealing interpretation (the conservative path), an explicit TDM exception (the EU-model path), and specific protections for individual creators against unauthorized commercial exploitation of their outputs in a way that resembles their distinctive work (the targeted path) — a combination that would permit training while guarding against specific harms. Underneath all of it is a view of copyright as a social bargain: limited monopoly rights granted to creators in exchange for eventual public benefit, with the bargain's terms calibrated to the social interest. Excessive restriction of AI training, in her analysis, would constrain a category of socially valuable activity (research, education, scientific work, accessibility applications, language preservation) to protect economic interests copyright wasn't primarily designed to protect. The framing question Craig poses: what is copyright for? Her answer leans toward "for the social and creative ecosystem broadly" rather than "for individual rights-holders maximally."
Where Geist and Craig converge. Both argue for an explicit TDM exception in Canadian copyright, against treating AI training as straightforwardly infringing, and for calibrating copyright policy to broader social purposes rather than to rights-holder maximalism. Both have the credentials and the published analyses to command serious consideration.
Where they differ from the TWUC and ACTRA positions. TWUC and ACTRA emphasize the individual-creator economic interest in compensation for AI training use. Geist and Craig emphasize the broader social interest in permissive frameworks for socially valuable AI applications. Both sets of values are legitimate. The disagreement is about how to weigh them.
The guide's methodological move is to refuse to resolve this disagreement. The chapter doesn't argue that Geist and Craig are right or that TWUC and ACTRA are right. It argues that the disagreement itself is the structurally important feature of the Canadian copyright-and-AI policy landscape, and that Canadian policy that doesn't engage both positions seriously is reducing a genuine intellectual debate to a political choice. **AIDA's failure, the absence of an AI-specific copyright provision in AI for All, and the deferral of the underlying questions to the courts and to future legislation, are all responses to the difficulty of resolving this disagreement.** Whether Canada's policy approach should be "let the courts resolve it" or "let Parliament make the choice explicitly" is itself contested. What's empirical is that no resolution currently exists in Canadian statute.
🧌 GOBLIN CHECK — The goblin's least favourite sentence in any policy fight is "so basically, one side is lying." Nobody in this chapter is lying. Geist and Craig are right that copyright maximalism has historically fattened aggregators more often than artists. TWUC and ACTRA are right that their members' work built these models and the cheque never came. Two true things, one statute. The hoard keeps both folders, cross-indexed, and the goblin refuses to shred either one just to make the filing easier.
---
The structural weakening of Canadian copyright
A specific structural finding deserves separate treatment because it underlies much of the rest of the chapter and connects directly to Chapter 20's path-forward analysis: Canadian copyright is not currently a comparatively strong jurisdiction for creator protections, and the AI moment is arriving in a regulatory environment that was already creator-unfriendly relative to peer countries.
The TWUC characterization, "severely weakened through Supreme Court of Canada decisions over the past 15 years and government inaction," points to specific identifiable factors:
The 2012 Supreme Court copyright pentalogy. Five Supreme Court of Canada decisions issued together in July 2012 that substantially expanded the fair-dealing exception in Canadian copyright law. The key holdings (in compressed form): fair dealing should be interpreted broadly rather than restrictively; user purposes (research, study, private review) qualify even when conducted in commercial settings; technological neutrality requires applying fair dealing flexibly across new technologies; aggregator services don't automatically become commercial infringers by enabling user fair-dealing activities. The decisions have been understood by Canadian copyright scholars as moving Canada substantially toward a US-style fair-use jurisdiction, though without the specific four-factor balancing test of US fair use.
The 2019 Copyright Act amendments and the post-2019 reform impasse. Following a parliamentary review of the Copyright Act, the federal government considered substantial reforms in 2019 but did not pass them. Subsequent reform efforts have similarly stalled. The 2024 ISED consultation on Copyright in the Age of Generative AI generated substantial expert submissions including the Geist, Craig, and TWUC positions discussed above — but did not result in legislative action. The federal government has chosen to engage AI copyright primarily through the failed AIDA framework rather than through Copyright Act amendments.
The Online News Act and its consequences. Canada's 2023 attempt to make platforms pay news publishers prompted Meta to block Canadian news rather than comply (the economics are Chapter 14's subject). The copyright-relevant lesson is the pattern: federal attempts to strengthen Canadian creator leverage against digital platforms have produced platform pushback rather than compliance, and AI-training litigation faces the same dynamics with even less leverage on the creator side.
Canada relative to peer jurisdictions. The EU has comprehensive copyright legislation including explicit AI-related provisions (the 2019 Directive's TDM exceptions and the 2024 AI Act's training-data transparency requirements). The UK has been actively legislating creator-side AI protections. Japan adopted an explicit TDM exception for AI training in 2018 (with creator-side concerns being raised more recently). South Korea is in active legislative engagement. The US has substantial case law developing, with mixed results: Andersen v. Stability AI favouring some creator claims; Bartz v. Anthropic splitting on fair use (training on purchased books yes, pirated copies no) before a reported US$1.5 billion class settlement in 2025 (final court approval still pending as of mid-2026); New York Times v. OpenAI still ongoing. Canada has minimal federal AI-specific copyright legislation, weaker general copyright protections than peer jurisdictions, and a court-by-court adjudication path that is slow, expensive, and uncertain in outcome.
This structural picture matters because it shapes the political-economy of the Geist/Craig vs TWUC/ACTRA disagreement. Geist and Craig's TDM-exception argument is operationally easier to implement because it leans on existing case law and aligns with international trends. TWUC and ACTRA's compensation-and-licensing argument is operationally harder because it requires either stronger legislation than Canada has shown the political will to pass, or judicial victories in cases that are slow and resource-intensive to litigate. The Geist/Craig path arrives at policy more readily; the TWUC/ACTRA path requires institutional capacity that has been eroded.
This is not an argument for which side is correct. It is an observation that the structural conditions of Canadian copyright shape which position is operationally more viable, and that ignoring those conditions in policy debate amounts to wishing for an outcome that the political economy doesn't support.
---
The bridge — NIL Rights, deepfakes, and the convergence question
A specific finding worth foregrounding because it connects this chapter to Chapter 13 (Misinformation & Deepfakes): the ACTRA NIL Rights argument and the CIGI likeness-rights-in-copyright argument are converging on the same legal proposal from opposite directions.
ACTRA argues for NIL Rights to protect performers against unauthorized AI-generated reproductions of their work in commercial contexts. The Centre for International Governance Innovation (CIGI), in its December 2025 analysis on deepfake legislation, argues for copyright-based likeness rights (following Denmark's proposed framework) to protect ordinary people against non-consensual deepfakes — sexual, political, or otherwise. Both proposals would establish in Canadian law a recognized rights claim over a person's likeness, name, image, and voice, enforceable against unauthorized commercial use.
The two proposals approach the same legal concept from different ends of the political spectrum. ACTRA's framing is creator-rights and commercial-economic. CIGI's framing is privacy-rights and harm-prevention. A single Canadian legislative reform, establishing likeness rights in copyright, could satisfy both proposals simultaneously.
This is the kind of convergent finding the guide's bias methodology surfaces. Two different constituencies, arguing for different reasons from different starting points, are pointing at the same policy mechanism. The federal AI strategy does not currently commit to this mechanism. Whether AI for All could be amended to include likeness rights legislation, whether new legislation should be introduced separately, or whether the existing fragmented approach (Bill C-16 for sexual deepfakes, ACTRA collective agreements for performer protection, common-law remedies for other cases) should continue — these are policy questions Chapter 20 returns to.
The convergence is structurally significant because it suggests a path through the Geist/Craig vs TWUC/ACTRA disagreement that doesn't require resolving the underlying philosophical disagreement. Geist and Craig can support TDM exceptions for general AI training while still supporting likeness rights against specific commercial exploitation of individuals. TWUC and ACTRA can pursue compensation frameworks for their members while still benefiting from likeness rights that protect ordinary Canadians. The convergent policy mechanism doesn't fully satisfy either position, but it advances both meaningfully. This is the kind of pragmatic policy synthesis that comprehensive AI legislation could include and that piecemeal regulation cannot.
---
The working test
It's the Chapter 3 test wearing a barrister's robe. For any AI copyright claim, ask: subject matter (what work, and who actually holds the rights: the creator, or a corporate aggregator using the creator's name to make the argument?); use (reproduction, derivative, training, output, commercial or not?); permission (licence, opt-out, fair-dealing assumption, or silence?); compensation (through what mechanism, if any?); statutory basis (section 3, section 29, section 30.71, or no clear footing?); jurisdiction (does Canadian law reach the use at all?); and verification (self-reported, or checkable?). Same method as every chapter: name what's happening, name the gaps, let readers decide.
---
The contest, left honest
CHAPTER RECAP — you now have: - The Canadian newspapers v. OpenAI case structurally understood — 16.1 million Canadian news works, Ontario jurisdiction confirmed November 2025, substantive trial pending, the case's reasoning will be precedential for Canadian AI copyright. - The TWUC and ACTRA positions in their own primary-source framing — TWUC on copyright infringement plus AI output ineligibility, ACTRA on NIL Rights plus Beijing Treaty implementation, both with specific operational mechanisms (model contracts, class actions, collective agreements, legislative proposals). - The Geist and Craig academic positions in their own framing — both arguing for explicit TDM exceptions in Canadian copyright, both grounded in different values (Geist on competitive Canadian AI development, Craig on copyright's social purpose), both rigorous and intellectually serious. - The structural finding that Canadian copyright is in a comparatively weak state relative to peer jurisdictions before the AI moment even arrives, and that the structural conditions shape which side's preferred policy is operationally more viable. - The convergent finding that ACTRA's NIL Rights argument and CIGI's likeness-rights-in-copyright argument point at the same legislative mechanism from opposite ends of the political spectrum, suggesting a path through the underlying disagreement. - The working test for evaluating any AI IP or copyright claim: subject matter, use, permission, compensation, statutory basis, jurisdiction, verification.
The next chapter (Chapter 13) takes the deepfake side of the personal-sovereignty discussion to its full treatment — Bill C-16's coverage and gaps, the 2025 Canadian election empirical study, the political-deepfake regulatory absence, the international comparison framework. The NIL Rights / likeness rights bridge introduced in Section 5 of this chapter gets its policy-mechanism analysis there.
You can now read any Canadian AI IP or copyright claim with the structural equipment to recognize whose position is being represented, what the underlying values are, and what the operational stakes are. The disagreement is genuine. The guide's methodology asks readers to hold the genuine disagreement, not to resolve it prematurely toward one side.
---
Bias label for this chapter: structural-political and academic-doctrinal analysis of Canadian copyright as it applies to AI, with explicit presentation of substantive academic disagreement rather than resolution toward one position. Author lean: methodological commitment to presenting genuine intellectual disagreements honestly; skeptical of "creators vs. AI" framings that flatten the genuine intellectual differences within both creator-side and AI-developer-side positions; willing to name the structural weakness of Canadian copyright as the context that shapes operational viability of different policy approaches; explicit that the manual does not attempt to resolve the Geist/Craig vs TWUC/ACTRA disagreement. Litigation parties (newspapers, OpenAI) treated as primary on the case itself. Creator-organization documents (TWUC, ACTRA) treated as primary on creator-side positions. Academic-doctrinal sources (Geist, Craig, the Martin-Bariteau and Scassa edited collection) treated as primary on the alternative-academic position. Government framing (ISED consultation, AI for All) treated as primary on what the federal government has committed to and has not committed to.
Primary sources cited or relied on in this chapter: Canadian newspapers v. OpenAI, Ontario Superior Court of Justice CV-24-00732231-00CL (November 2024-present); Writers' Union of Canada formal position page and January 2024 ISED consultation submission; Alliance of Canadian Cinema, Television and Radio Artists, 2024 AI Submission Explainer and 2023 internal member survey; Independent Production Agreement ratified January 21, 2025; Directors Guild of Canada and Music Canada parliamentary testimony on Bill C-27 (February 2024); Carys Craig, 2024 ISED submission (SSRN 4718941); Carys Craig, chapter in Martin-Bariteau and Scassa, AI and Copyright (2021); Michael Geist, January 2024 ISED consultation submission and subsequent academic writing; Centre for International Governance Innovation analysis on deepfake legislation (December 2025); 2012 Supreme Court of Canada copyright pentalogy (CCH Canadian, Society of Composers, Alberta (Education), and related cases); Beijing Treaty on Audiovisual Performances (2012 WIPO treaty); European Union Digital Single Market Directive (2019); Online News Act (Canada, 2023). Detailed citations in the Sources appendix.
---
🧌 GOBLIN CHECK — The goblin's least favourite sentence in any policy fight is "so basically, one side is lying." Nobody in this chapter is lying. Geist and Craig are right that copyright maximalism has historically fattened aggregators more often than artists. TWUC and ACTRA are right that their members' work built these models and the cheque never came. Two true things, one statute. The hoard keeps both folders, cross-indexed, and the goblin refuses to shred either one just to make the filing easier.
Recap
- The Canadian newspapers v. OpenAI case structurally understood — 16.1 million Canadian news works, Ontario jurisdiction confirmed November 2025, substantive trial pending, the case's reasoning will be precedential for Canadian AI copyright.
- The TWUC and ACTRA positions in their own primary-source framing — TWUC on copyright infringement plus AI output ineligibility, ACTRA on NIL Rights plus Beijing Treaty implementation, both with specific operational mechanisms (model contracts, class actions, collective agreements, legislative proposals).
- The Geist and Craig academic positions in their own framing — both arguing for explicit TDM exceptions in Canadian copyright, both grounded in different values (Geist on competitive Canadian AI development, Craig on copyright's social purpose), both rigorous and intellectually serious.
- The structural finding that Canadian copyright is in a comparatively weak state relative to peer jurisdictions before the AI moment even arrives, and that the structural conditions shape which side's preferred policy is operationally more viable.
- The convergent finding that ACTRA's NIL Rights argument and CIGI's likeness-rights-in-copyright argument point at the same legislative mechanism from opposite ends of the political spectrum, suggesting a path through the underlying disagreement.
- The working test for evaluating any AI IP or copyright claim: subject matter, use, permission, compensation, statutory basis, jurisdiction, verification.
Sources
- Canadian newspapers v. OpenAI, Ontario Superior Court of Justice CV-24-00732231-00CL (November 2024-present)
- Writers' Union of Canada formal position page and January 2024 ISED consultation submission
- Alliance of Canadian Cinema, Television and Radio Artists, 2024 AI Submission Explainer and 2023 internal member survey
- Independent Production Agreement ratified January 21, 2025
- Directors Guild of Canada and Music Canada parliamentary testimony on Bill C-27 (February 2024)
- Carys Craig, 2024 ISED submission (SSRN 4718941)
- Carys Craig, chapter in Martin-Bariteau and Scassa, AI and Copyright (2021)
- Michael Geist, January 2024 ISED consultation submission and subsequent academic writing
- Centre for International Governance Innovation analysis on deepfake legislation (December 2025)
- 2012 Supreme Court of Canada copyright pentalogy (CCH Canadian, Society of Composers, Alberta (Education), and related cases)
- Beijing Treaty on Audiovisual Performances (2012 WIPO treaty)
- European Union Digital Single Market Directive (2019)
- Online News Act (Canada, 2023).