When I appeared before the Standing Committee on Canadian Heritage last fall for its study on AI and the creative industries, I emphasized that the large language models and generative AI systems that are reshaping how people access information, culture, and entertainment are only as representative as the data on which they are trained. If Canadian works, perspectives, and cultural content are absent from those models, Canada risks being diminished in the AI-mediated world. The committee’s report, released this month, acknowledges this concern, but its lead recommendation would make the situation worse.

The report’s centrepiece is Recommendation 1(c), which calls on the government to “establish a clear opt-in consent requirement for the use of copyrighted works in the training of artificial intelligence systems, ensuring that creators’ works may not be used for text and data mining or model development without their prior authorization.” In other words, the committee wants to stop the use of copyrighted Canadian works for AI training unless every rights holder has given advance permission leaving Canada as an outlier compared to peer jurisdictions.

For example, the European Union, often regarded as the global leader for cultural protection and creator rights, operates on an opt-out model: under Article 4 of the Directive on Copyright in the Digital Single Market, text and data mining is permitted unless rights holders expressly reserve their rights in machine-readable form. Japan’s Copyright Act goes further, permitting the use of copyrighted works for data analysis purposes, including AI training, without any opt-out mechanism, subject to a caveat guarding against unreasonably prejudicing copyright owners. Singapore’s Copyright Act includes a specific exception for computational data analysis, including for commercial use. The Israeli government has issued an opinion stating that AI training is generally permitted under existing fair use provisions, an approach consistent with the views of many in the United States.

The practical consequence of the Heritage recommendation is easy to predict, because we have seen it before. When the Online News Act created conditions requiring platforms to pay for news links, Meta responded by blocking news content in Canada entirely. As I mentioned to the committee, erecting barriers that impose new costs on the inclusion of Canadian content in AI models risks the same outcome. Rudyard Griffiths, publisher of The Hub, made the same point more bluntly, calling the Online News Act a “cautionary tale” and noting that publishers who participate in the Google deal are now required to make all of their content available for scraping, including behind paywalls, as a condition of receiving funding.

The opt-in framework also amounts to de facto copyright reform, though the committee avoids saying so directly. Canadian copyright law includes fair dealing provisions that allow the use of copyrighted works for purposes such as research and private study. While there are benefits to providing greater certainty, most TDM for AI training purposes would likely qualify as fair dealing under existing law. A mandatory opt-in requirement not only rejects a new TDM exception but also effectively overrides fair dealing by imposing a prior-authorization requirement not found in the existing law.

How did the committee arrive at a recommendation so at odds with the competitive positioning it claims to promote? The witness list tells much of the story. Over seven meetings, the committee heard from 43 witnesses, the overwhelming majority of whom represented collective rights organizations (Access Copyright, Copibec, SOCAN, Music Canada, Music Publishers Canada), cultural industry associations (CMPA, Directors Guild, Writers Guild, Union des Artistes), and cultural policy advocates. The handful of witnesses who offered a different perspective on copyright and competitiveness, including AI researchers, the entertainment software sector, and those who cautioned against overregulation, were cited in the report but functionally absent from its recommendations. The committee then presented its conclusions as if they reflected a consensus, adopting the “ART” framework (Authorization, Remuneration, and Transparency) promoted by the collective rights organizations as its organizing principle. The reality, however, is that the testimony “did not yield a consensus on the exact means of striking the balance between facilitating innovation and protecting Canadian creators.” That’s not my line, that’s a quote from the committee report.

This pattern is not new. As I noted when the government released the results of its AI consultation earlier this year, there is a recurring tendency to manufacture an illusion of consensus by softening genuine disagreements among experts into government-friendly summaries. The Heritage committee report follows the same playbook: the body text captures real tensions and competing views, while the recommendations resolve every tension in a single direction.

The question the government must confront as it considers this report is whether it wants more Canada or less Canada in the AI systems that are increasingly mediating how the world encounters information and culture. An opt-in consent requirement, applied to a country whose cultural output is a fraction of the global training corpus, is a policy designed to produce less. There is plenty of opportunity for market-based deals in the current environment, but if the government follows this recommendation, it risks ensuring that Canadian cultural content is absent from the AI models that matter most.