Libraries Are Complaining About the Cost of Licensing E-Books (Again): Is There Any Justification?

An e-book reader displaying the text 'E-BOOK What book you will read today?' next to a stack of hardcover books and an open book.

Image: Shutterstock.com

Is it a coincidence that a story complaining about the cost of e-books in the budget of the Greater Victoria Public Library (GVPL) appeared in my local paper, the Times-Colonist, earlier this month only to be followed mere days later by an almost identical story that popped up in a news scan service I receive, but this time relating to the Washington, DC, Public Library? I suspect not. In the case of the GVPL, the library has to shell out $57.00 for a two-year licence for an e-book that costs $14.99 for an individual licence. Both the Canadian Library Association and its US counterpart, the American Library Association, have had this issue on their agenda for a few years now. The Canadian Urban Libraries Council, which represents more than fifty of the largest public library systems in Canada, has a Digital Content Working Group that advocates for “equitable, affordable, and sustainable access to digital content for public libraries across Canada.” Who could argue with that? But there is always a devil hiding in the detail.

The major difference between the two stories was that the GVPL was simply lamenting the budgetary impact of licensing e-books, which are in popular demand from borrowers, whereas the US story not only raised the cost issue for libraries but also referred to a local legislative initiative that would authorize market intervention by imposing restrictions on the District’s public library system, preventing it from licensing e-books from publishers that reportedly charge “excessive prices”. But what is excessive and what is reasonable when it comes to e-books, and can or should governments intervene in the market to regulate the price of a discretionary product like a book?

The DC legislation has not been adopted and if it passes, it may still never go into effect since it contains a proviso that implementation will take place only once ten other jurisdictions with a combined population of at least 50 million have enacted substantially similar laws. This is apparently to protect the DC Council from lawsuits and to try to build a national coalition to negotiate collectively with publishers. It also avoids the very real possibility that publishers could opt to not license e-books to the District of Columbia Public Library system. The bill would actually hurt libraries and their patrons by prohibiting the DC library system from entering into and renewing licensing agreements with publishers if those agreements contain terms that “restrict public access to books.” What does “restricting public access” mean? Are publishers not legally entitled to determine the terms under which their product is offered to various categories of users? Should publishers be legally required to license their product on terms arbitrarily set by government? Most licensing agreements, which are contracts, impose conditions governing how a licensed product can be used by different categories of users, such as (for a library) the number of times it can be loaned and how long the agreement lasts before it lapses and must be renewed. These terms are set by the entity offering the product, i.e. the publisher. There is no obligation on any prospective licensee to accept the terms. Products that are priced beyond what the market will accept will not succeed. The argument of the library community seems to be that libraries are somehow different, a public good, and therefore they should pay what individual consumers pay even though their use of the product is quite different.

The outline of the DC bill states that restricting public access includes “inflating” costs beyond what the public pays. Under standard licensing terms for digital books offered to the public, aka Joe and Jill Consumer, they pay less than libraries do for access to the same work. This has been the case for e-books since the early days of digital content, and it is not difficult to understand why. When a library buys a hard copy book, which it can do for the same price as the consumer (in fact, usually for less since library systems buy in volume and thus enjoy discounts), it can expect to lend that book for a limited period before it wears out and a replacement has to be purchased. There are other risks as well when hard copy works are circulated on loan, such as loss and damage. Moreover, while he majority of borrowers (in Victoria it is 55%) still generally prefer to borrow a hard copy version, there is “friction” in borrowing a physical book that limits the number of times a work is taken out. This friction is the unavoidable time and cost of physically accessing a library, such as parking or transit costs, deterrence from inclement weather or just sheer laziness. E-books, on the other hand, can be accessed with a click of a mouse. Each copy is perfect. No dog-eared or missing pages here. No borrowing “friction”.

While “friction” is an undesirable byproduct of the traditional library system, imposing constraints on borrowing physical products for some, from a commercial perspective it imposes a reasonable limit on use. The absence of “friction” where digital products are concerned not only makes things easier for the borrower, but it also reduces costs in other areas for libraries. E-books require virtually no physical storage space, and while some clients will go to a library to access an e-book, many do so remotely, thus reducing physical visitorship and the need for extended hours, etc. One of the key arguments put forward by publishers to justify the higher cost of e-book licences to libraries is that individual consumers cannot circulate digital copies to others (unless they physically lend the device on which their e-book is loaded) whereas lending to multiple users is the essential function of a library. An e-book licensed by a library for lending to multiple users is, in effect, a different product from the single-use product licensed to an individual consumer. Moreover the “perfect-every-time” perennial nature of an e-book eliminates the need for any book replacement over time. Combined with potential savings gained in other areas, the pleas of public libraries to have legislators impose restrictions on publishers have to be seen in context. That said, I am sure this will not stop library associations from trying to convince legislators to tilt negotiations in their favour.

The DC Council is modelling its legislation on a similar law passed last year in Connecticut. Like DC’s draft law, the Connecticut legislation will come into effect only when another state or states totalling seven million in population enact similar laws. For the record, Connecticut has a population of about 3.6 million people. Neighbouring New York state has a population of 20 million so if it goes the same route, the ball will start rolling. The caution of both DC and smaller jurisdictions like Connecticut is understandable because one response from publishers to legislation that restricts their ability to offer industry-standard contracts/licensing agreements could be non-availability of e-books for libraries in the affected areas. Libraries need to be careful what they wish for, since a heavy-handed approach by legislators could backfire.

Connecticut claims its legislation avoids the legal pitfalls that befell a 2022 law passed in Maryland that was blocked, on appeal from the American Association of Publishers and the Authors Guild, because the state law impinged on US federal copyright law. That state law required authors and publishers holding the rights to an e-book title to, among other things, offer unlimited copies of that title to public libraries in the state at an undetermined “reasonable price” when the title was offered to individual consumers. In effect it was tantamount to imposing a compulsory license on rightsholders and was rightly blocked by the courts. I wrote a long blog post on the e-book licensing situation at that time. You can read it here.

While the Maryland approach was overreach, the issue hasn’t gone away. Library budgets are under stress but that should not be used as an excuse to force subsidization from authors and publishers. As mentioned above, e-books never wear out and a widely circulated “frictionless” library e-book is not a direct substitute for the hard copy work that libraries used to buy–and still buy but in lesser amounts. It is a different product and to compare it to a hard copy book is to ignore commercial realities. While libraries are a public good, they still operate in the marketplace and affect the market for books, in some ways positively and in others negatively. Governments normally put their thumb on the scale of commercial negotiations only when there is market failure. This is far from the case today. The case for market intervention by government is weak, although I sympathize with the plight of acquisition librarians facing stretched budgets and trying to meet the increasing demand for e-books.

There are market limits to what publishers can charge libraries for e-books just there are market limits regarding what public libraries can pay. What is the appropriate licensing fee for an e-book configured for library use versus one sold to the individual reader? Is a multiple of three or four justified, or should it be less (or more?). An objective study would be helpful. Let’s hope there is room for negotiation that will fairly compensate authors and publishers while allowing public libraries to meet the growing demand of their clients for more digital works.

© Hugh Stephens 2026. All Rights Reserved

Canadians (and Anyone Else Outside the US): Beware the Annual Public Domain Hype

A black and white cartoon character resembling a mouse, wearing a hat and shorts, happily steering a ship's wheel.

Image: Public Domain

This is the time of year (the days and weeks after January 1) when, on a quiet news day, lazy journalists in Canada used to pick up and amplify a US based story about such and such a work falling out of copyright and into the public domain and write a story about it, complete with a grabbing headline, often to the effect that Mickey Mouse or Batman or The Great Gatsby or whoever is now “liberated from the chains of copyright”. The CBC did exactly that in 2024, producing a radio special on Steamboat Willie entering the public domain, completely ignoring the fact that all works created by Walt Disney and Ub Iwerks (which include the earliest editions of Steamboat) had already entered the public domain in Canada two years earlier, on January 1, 2022. Iwerks was co-author and joint rightsholder along with Disney for this work and as the co-author who lived the longest (he died in 1971; Disney in 1966) the term of copyright protection in Canada was based on the year he died plus, at the time, an additional 50 years. Thus, January 1, 2022 in Canada. As I wrote a couple of years ago (Canada is not the United States when it comes to Copyright: The Cases of Anne of Green Gables and Steamboat Willie (or Down the Copyright Rabbithole, Twice), sometimes works still under copyright protection in Canada are in the public domain in the US, and sometimes it is the reverse. Don’t assume. This law firm’s blog post (Gowlings) provides a good overview of what to watch out for.

The mistake of making the assumption that what happens in the US is automatically applicable to Canada is the unfortunate reality of being a cultural minnow living cheek by jowl with a content creation whale. This year I didn’t notice any of the reflected US public domain stories in the Canadian media, perhaps because the penny has finally dropped that public domain day in Canada will be a non-event until the year 2043 (no need for any hype), owing to the extension of Canada’s term of copyright protection from the life of the author plus 50 years to life plus 70 (for most works). Any works that had fallen into the public domain under Canada’s previous “life plus 50” term did not receive the additional term of protection but any works still copyright protected in Canada at the end of 2022 got another twenty years coverage before entering the public domain. Had Ub Iwerks died in 1972 instead of a year earlier, Steamboat Willie would have enjoyed another two decades of protection in Canada beyond what applies in the US, although it is doubtful whether the Walt Disney Company would have tried to enforce its rights under Canadian law.

Not only has Canada harmonized its current term of copyright protection with the US, EU, UK, and a number of other countries, (although there will always be discrepancies between the terms of protection afforded works in Canada versus those in the US for many years to come owing to the historical peculiarities of how the term of protection is calculated under US copyright law), there have also been fewer quiet news days in early 2026 thanks to the daily Donald Trump Reality Show. Moreover, there has been a surge in Canadian nationalism (and thus a greater awareness of cultural differences) as a result of the Donald’s 51st state taunts. (Prime Minister Mark Carney, an internationally recognized banker and financial executive seemed initially to enjoy Trump’s respect but since his Davos speech calling out the realities of the new world order, Carney, like his predecessor Justin Trudeau, has been demoted to the title of “Governor Carney” on Truth Social, apparently the current official channel for announcements of US government policy). So, journalists, if you want to write about what makes Canada different from the US, in addition to measuring distance in kilometers and saying “sorry” every time someone bumps into you, you could note that US and Canadian copyright laws are different. Similar in intent but not identical. For example, the US fair use doctrine with its unpredictable focus on transformative use does not apply in Canada, the US requirement for formal registration of copyright in order to bring legal action does not apply in Canada and, in particular, the complex (because of its convoluted history) US determination of when a particular work falls into the public domain does not apply in Canada.

This year, as it does every year, the Center for the Study of the Public Domain at Duke University’s Law School, published its Public Domain Day blog, highlighting all the works that fell into the US public domain on January 1, 2026. These include such well known works or characters as Agatha Christie’s The Murder at the Vicarage (protected in Canada until January 1, 2043) , Somerset Maugham’s Cakes and Ale (in the public domain in Canada since 2016), Blondie and Dagwood, nine additional Mickey Mouse cartoons, Dutch artist Piet Mondrian’s Composition No. II/Composition in Red, Blue, and Yellow, four songs by Ira and George Gershwin, and so on. Much is made of the fact that these works will be free to anyone to use, remix, copy and exploit but it’s not as if these works have been locked away in a closet, although their unlicensed use has been protected by copyright law. Copyright protection does not stop anyone from creating new works and while there may be limitations on hijacking Inspector Poirot there is nothing stopping aspiring writers from creating detective novels. There is another element worth noting as well. Particularly when it comes to copyrighted characters and cartoons, their later iterations may still be copyright protected in the US (because of the US baseline being date of publication plus a set number of years) not to mention protection offered by registered trademarks. Sometimes estates try to hang on to copyright protection at all costs, as appears to be happening in the US with Mondrian’s work. (It has been in the public domain in Canada since 1995, 50 years after Mondrian’s passing).

Once a work has entered the public domain it can be used in derivative works without permission. Has this resulted in a slate of new and creative works being produced for the benefit of mankind? Hardly. The usual result is for a brief surge of “edgy” productions incorporating a new public domain work, such as Steamboat Willie doing or saying things that Dear Old Uncle Walt (Disney) would never have countenanced. As I noted a couple of years ago, the “liberation” of copyright protected works has led to such triumphs as The Gay Gatsby, The Great Gatsby Undead (Zombie Edition) and the film Winnie the Pooh: Blood and Honey. So much for the public domain unleashing the juices of creativity.

For better or worse, copyright is not a perpetual property right. I support reasonable limitations on copyright protection including making a provision for works to enter the public domain after their prime exploitability has passed. This will vary by work with some works remaining evergreen, encouraging new investment into derivative works, updates, new editions, as well as providing ongoing returns to the estates of authors. However, at times there are situations where a work is long out of print and the rights-holder cannot be located, blocking a reprint. These situations can be dealt with through specific exceptions, much as fair use and fair dealing allow for specified unauthorized uses that do not damage the rights of the author.

To come back to the narrative that the public domain liberates content from the “shackles of copyright”, I contend this is nonsense. Beware the hype. And if you reside outside the US, don’t believe everything you read in the media regarding what works are in the public domain. You might be pleasantly surprised to find that a work has been in the public domain in your country for years (while Conan Doyle’s later works only entered the US public domain in 2023, they have been in the public domain in Canada since 1981). On the other hand, you might find the work you thought was free for adaptation based on what Duke University’s Center for the Study of the Public Domain says is way off base and it is still protected by copyright in your country of residence. Beware the hype and do your homework.

© Hugh Stephens, 2026. All Rights Reserved.

India’s Proposed Compulsory Licensing Scheme for AI Training: It Could be the Worst of All Worlds.

A wooden gavel striking a golden copyright symbol with the text 'COMPULSORY LICENSING' in bold.

Used with permission

In early December the Indian government through the Department for Promotion of Industry and Internal Trade (DPIIT), launched a public consultation on its proposal for a “One Nation, One Licence, One Payment” regime to govern AI training on copyrighted content. The comment period closes early next month. The stated intent is to establish a framework to allow AI developers to access proprietary content for AI algorithm training, without rightsholder authorization, while ostensibly taking into account the concerns and economic interests of those same rightsholders. The proposal from a semi-official committee established by DPIIT is based on a compulsory licence regime covering all content, past, present and future, with no opt outs for content owners. Given the unrepresentative nature of the committee, which failed to include rightsholders from many creative sectors, it is not surprising it has managed to come up with a solution that pleases absolutely no-one, not rightsholders nor the AI industry. Moreover, it is probably unworkable and could rightly be described as the “worst of all worlds”. How did we end up here?

It is widely accepted that to further refine and develop AI, more data is constantly required. Quality data is important. The better the data, the better the product. Many AI developers have, without authorization, already helped themselves to vast amounts of copyright protected material either through accessing pirate databases or by simply hoovering up publicly accessible (but nonetheless copyright protected) content on the internet. The result has been a series of lawsuits launched by rightsholders, primarily in the US but also in India, the UK and Canada. In the US the parameters of the fair use doctrine are being tested, with mixed results. There have been some settlements, notably the Bartz v Anthropic case in which AI developer Anthropic agreed to pay a group of authors $1.5 billion for having accessed and reproduced their works through pirate websites (and without authorization), as well as a large number of voluntary licensing agreements between corporate rightsholders such as media outlets, publishers and music labels, and AI companies. Clearly, legal leverage is needed to convince AI developers to negotiate with rightsholders (no one is volunteering, that’s for sure). In the US this is backed up by a statutory damages’ regime where the risk of being hit with large statutory damages per infraction is a strong incentive for the AI industry to reach licensing deals. Unfortunately, many countries do not have a statutory damages provision in their copyright law, and India is one of them.

Ironically, outside the US where the fair use case-by-case legal doctrine does not exist, AI developers are arguably in even greater legal jeopardy. As a result, they have pushed for a wide statutory text and data mining (TDM) exception to be introduced into national copyright laws. The AI industry is doing this in India, although there is strong opposition to introducing a TDM escape hatch. When a TDM exception applies, rightsholders are paid nothing. The DPIIT “solution” proposes to address the non-payment issue–but is going about it in precisely the wrong way. Appropriating the IP of all rightsholders by imposing a draconian compulsory license regime penalizes rather than rewards rightsholders and imposes a lowest-common-denominator value on all content, not distinguishing between premium and pedestrian content while taking away from rightsholders the inherent right to determine how and where their content is used.

The compulsory licence regime proposed by DPIIT would be administered by a new non-profit body, a Collective Management Organization (CMO) to be established by statute. This would create another layer of bureaucracy harking back to the days of the “licence raj”. A government appointed committee would set rates in conjunction with the new non-profit. Royalties would be applied retroactively as well as prospectively. Existing voluntary licensing agreements between content providers and AI companies would have to be terminated and future licence agreements prohibited. Rightsholders who do not want to licence their content could not opt-out. This is overreach at its worst.

The tech industry represented by NASSCOM, the National Association of Software and Service Companies, that also includes Google and Microsoft, vehemently opposes the One Nation, One Licence, One Payment proposal, arguing instead for a TDM exception. It also argues that rightsholders who wish to opt out should be able to do so, thus avoiding AI companies having to pay royalties for content it does not want or has already licensed. It also doesn’t want the precedent of compulsory payments, as India would be the first country globally to institute such a regime. NASSCOM claims that a compulsory licence regime would slow innovation, the card always played by the AI industry when the issue of protecting rightsholders is discussed. However, a TDM exception is not currently on offer in India. It is a “solution” that is facing increasing pushback in many countries. Australia has just rejected the concept, and other jurisdictions are examining it critically.  Where a TDM exception does exist, as in the UK and EU, its use is constrained and subject to several conditions, such as in Britain where it is limited to non-commercial research.

Not only is the tech industry strongly opposed to the compulsory licence proposal, so are rightsholders such as the broadcasting industry and Bollywood. Not only would a compulsory licence be extremely difficult to implement given the nature of CMOs which are generally highly inefficient and administratively cost heavy– not to mention that the Committee’s working paper proposes two layers of CMO (thus double handling and processing)– but the draconian and sweeping nature of the compulsory licence is of great concern. Rightsholders are given no ability to opt out or to refuse to have their content conscripted. In fact, the proposal includes a provision granting AI developers access to content as a “matter of right”. What happened to the fundamental right of an author to determine how, when and even if their content is to be reproduced? Compulsory licences are extraordinarily blunt instruments and do not work when sophisticated content like audio-visual and music products are involved. There are many elements to licensing, including contractual issues that go beyond price that are carefully negotiated and carry with them specific obligations and privileges. A compulsory licence is a “one size fits all” solution. It strips away the rights of content owners and is, in effect, a form of “compensated expropriation”. And the compensation is minimal.

Having pleased no-one with its proposal, it remains to be seen where DPIIT will go next once all comments are received and evaluated. A follow-up proposal is expected that will deal with outputs, just as the initial one did with inputs. India, like many countries, wants to participate in the global development of AI. Its rich local-language content is a strategic asset. Imposing a compulsory licence would stifle the creativity that drives this cultural comparative advantage. In the absence of a market failure there is no rationale for resorting to the sledgehammer of a compulsory licence for all content. The preferred solution for rightsholders, voluntary licensing, is also becoming a preferred solution for AI developers as the legal ground on which they are operating in accessing content without authorization is looking increasingly shaky. Voluntary licensing is growing globally as AI developers absorb this reality. If wide TDM exceptions are off the table, AI developers will have no recourse but to negotiate licence agreements. (A statutory damages regime would provide even greater incentive to do so). Bringing in a wide TDM exception or worse, introducing a blunt and highly bureaucratic compulsory licence regime, is a sure way to kill the growing voluntary licensing market.

DPIIT’s next steps will be crucial. If the goal is to create the conditions for a “made in India” AI industry while nurturing and protecting India’s valuable cultural assets there is no better solution than to create the conditions for the growth of a voluntary licensing market. Strong cultural industries and robust AI development go hand-in-hand. This means dispensing with forced, bureaucratic solutions like the proposed compulsory licensing regime while holding firm on rejecting a TDM loophole that would allow AI companies to plunder India’s cultural richness without any compensation to creators and rightsholders. Let’s hope India gets it right.

© Hugh Stephens, 2026.  All Rights Reserved

The Online Streaming Act or Dairy Supply Management: Which one should Canada Surrender to the US in CUSMA Trade Negotiations? Or is it a Question of Putting Some Water into the Wine of Both?

A person pouring water from a ceramic jug into a large clay pot.

Image: Shutterstock

Should Canada give up the Online Streaming Act (OSA) in forthcoming CUSMA negotiations in order to preserve dairy supply management, as a former Vice Chair of the CRTC, Peter Menzies, suggested in a Globe and Mail oped earlier this month? Perhaps he was just being deliberately provocative although the question hits one of the raw nerves of Canadian politics. The cultural community– particularly in Quebec–would be enraged if this happened. But then if supply management is watered down to allow more imports from the US, especially in dairy, the dairy farmers–particularly in Quebec–will be equally enraged. Which group has the greater political clout? In both cases, Quebec-based interest groups have a card to play denied to others in Canada. It is called the Bloc Quebecois, and if enough support bleeds from the Liberals to the Bloc, that could just open the way to the Conservatives to form the national government they so desperately crave. The cultural mavens in Toronto have little choice; either support the Liberals or face a worse fate when those Conservative cowboys from Alberta take the reins of power.

The Quebec cultural community which insists that measures are needed to ensure that foreign streamers both contribute financially to support Canadian content (Cancon) and ensure that Cancon (when expressed in French) is “discoverable”, has yet another card up its sleeves. It is called Bill 109, Quebec legislation (that is probably ultra vires since broadcasting clearly falls within federal jurisdiction) that purports to regulate the discoverability of French-language cultural content in the digital environment. If Canada gives way on the Online Streaming Act in CUSMA negotiations, watch Quebec fill the void. So where does all this leave the Carney government? Between a rock and a hard place.

It is true, as Menzies has pointed out, that the CRTC has been very slow, plodding even, in dealing with implementation of the OSA. It may even be overwhelmed, with inadequate staff as he suggests. The fact that implementation is still a work in progress makes it easier for the US government to bring pressure to stop or at least to modify rollout of the legislation, whereas other objectives mentioned in recent USTR hearings, such as changes to the Bank Act to benefit US financial institutions or measures to terminate supply management would require significant legislative and regulatory change to undo measures that have been in place for decades. Best to nip it in the bud, or to kill it in the egg, as they say in Quebec.  

Canada’s planned introduction of a Digital Services Tax (DST) is a prime example of a nipped-in-the-bud policy. A DST deals with tax avoidance measures implemented by large digital multinationals by taxing their in-country revenues rather than their manipulated profits. Some countries, such as the UK, France, Spain, Italy etc had already implemented a DST before Trump’s return to office and seem to have got away with it, even though Google, Microsoft, Amazon, META and others of that ilk have the Trump Administration’s ear. Canada intended to implement a DST several years ago but dithered and dragged its feet, finally passing legislation in 2024 that would have brought a DST into effect on June 30, 2025, backdated to 2022 when the law should have been put into effect in the first place. Unfortunately for Canada, the implementation date fell right in the middle of the Trump tariff war and Canadian efforts to negotiate some relief. But rather than postponing implementation yet again–and using the possibility of a future DST as negotiating collateral–Canada “bravely” announced it was going ahead with implementation (regardless of the consequences). Until it wasn’t. Trump tweeted that he was cancelling trade negotiations with Canada because of the DST and voilà, over a weekend, the DST was cancelled (on June 29, 2025).

Trade talks resumed and actually appeared to be making some progress with respect to sectoral tariffs such as steel until the next excuse Trump found to end them. This time it was over Ontario’s World Series free trade ads that ran on US television, using Ronald Reagan’s words from a 1980s era speech praising free trade and condemning protectionist tariffs. The content of those ads may have been accurate, but the result was one of Canada’s more prominent “own goal” moments. While Doug Ford may have derived some brief satisfaction from getting under Donald Trump’s skin, the steelworkers of Sault Ste. Marie, who might have benefited from a rumoured sectoral deal on steel, have been paying the price. I think this fiasco helps explain the public anger of US Ambassador to Canada Pete Hoekstra (who surely wins the 2025 “Bull in a China shop” award) who crudely vented his frustration that a deal so close to fruition got blown out of the water through Premier Doug’s ill-considered initiative.

But what about supply management? Canada should be taking a long, hard look at the wisdom of continuing to defend this 1970s policy that almost every other country has since abandoned. Instead, it should use the CUSMA negotiations as the reason to ditch a monopoly that protects a few chosen producers of supply managed commodities at the expense of consumers and the rest of the economy. Unfortunately, that won’t happen because of Canadian political realities but there is still scope for some wiggle room. In recent years, Canada has been forced as part of its trade negotiations to open slivers of the dairy market to EU countries, CPTPP trading partners, and to the US through the CUSMA. The dairy industry screamed blue murder but was paid off for having to face a bit more competition. As part of liberalizing as little as possible, Canada routinely plays games with its commitments by awarding import quotas to the same domestic dairy industry with which exporters of dairy products to Canada are competing. Some additional foreign cheese and dairy products become available to consumers but in effect the fox is in charge of deciding which chickens get let in, and at what price. Even though this policy is an albatross around Canada’s neck, such is the power of the dairy industry (which is reputed to control the outcome of no less than eight ridings in Quebec) that all political parties support keeping supply management off the table in all trade negotiations, and passed legislation to this effect. In a political environment where the government is one vote short of a majority, risking the ire of Quebec dairy farmers is a risky business.

Does that mean that supply management is completely off the table and instead there should be another sacrificial lamb, such as domestic broadcasting and cultural policy, as Peter Menzies has suggested? This is a doubtful proposition. Despite all the posturing about supply management being “off the table”, there will almost certainly be some concessions to the US, even if it is only in the way the tariff free import quotas are managed. The Carney government will claim it is defending supply management, while making some tweaks to the system. It can do the same for cultural industries. Defend the essence but find compromises that US industry can live with.

Like supply management, the Online Streaming Act also has wiggle room in its implementation. Already we have seen the CRTC announce changes to Cancon definitions that introduce greater flexibility and go some way toward meeting the concerns of the (largely US-based) content streamers, while preserving elements of protection for Canadian production. (Canadian makeup and hair design artists will be happy as use of their services adds an element of “Canadianness” to a production that could be useful in meeting the Cancon definition. This just goes to show that you can never discount the influence of a specific lobby). While the US has laid out some maximum wish-list objectives, including withdrawal of the Online Streaming Act (as well as the Online News Act), there are domestic political realities in Canada that will constrain Canadian trade negotiators from sacrificing the cultural sector to gain other objectives, just as there are with regard to supply management. The US may hold a big stick in the negotiations, but Canada is not without cards to play. It just has to be careful how to play them, and when mobilizing support inside the US to do so in a way that does not offend the touchy amour-propre of Donald Trump.

The end result for Canada will not be water or wine, but rather how much water to allow into the wine. Some dilution will be necessary but at the end of the day, for domestic political reasons (particularly in Quebec), the liquid in the glass still will still have to resemble wine more than water. This applies equally to cultural industries and broadcasting as well as to supply management. It is far from an either/or situation.

© Hugh Stephens, 2026. All Rights Reserved.

Vietnam’s New IP Law: The General Direction Makes Sense but Regulatory Details will be Key

Three musicians perform on stage at night, surrounded by colorful decorations and ambient lights.

Image: Shutterstock

On December 10, Vietnam’s National Assembly adopted an amended IP (intellectual property) law designed to update the legislation for the digital age, and especially to address issues involving the intersection of AI and IP. The next phase will be drafting of the regulations necessary to bring the law into effect by April 1, 2026. The law does several things; it will explicitly recognize IP as an economic asset allowing it to be used for financial and accounting purposes, and as collateral for loans; it clarifies that works autonomously generated by AI without substantive human creative input are not eligible for IP protection; and it allows the use of lawfully published works and data for AI research, testing, and training, provided that such use does not unreasonably prejudice the legitimate interests of rights holders and the resulting outputs do not infringe copyright. While all this sounds good on the surface, there are potential hidden problems with the AI research, testing and training provisions. In drafting the implementing regulations, Vietnam needs to be careful to not undermine its vibrant creative industries that depend on copyright protection to thrive.

The translated wording of the AI provision (Clause 5, Article 7) is rendered in English as follows:

“Organizations and individuals may use text and data related to intellectual property rights that have been lawfully published and made publicly accessible for purposes of scientific research, experimentation, training of artificial intelligence systems, provided that such use does not unreasonably affect the legitimate rights and interests of the author or the holder of intellectual property rights as stipulated by this Law.

For text and data that are subject to copyright and related rights protection, the use of such text and data under this Clause must also comply with the Government’s regulations.”

At first glance, this looks like another “standard” Text and Data Mining (TDM) exception, a carve out from copyright protection being pushed by the AI industry in a number of Asian countries, as I recently wrote about here. In that blog post, I noted that lawmakers and regulators in Asia are grappling with a common problem; how to incentivize the development of responsible AI while continuing to encourage and promote all-important content industries. Vietnam is facing the same issue. Throwing the cultural sector under the bus in the hopes of attracting some ephemeral hi-tech AI jobs is a false bargain. Strong cultural industries enable the development of strong content licensing markets for AI development, enabling a virtuous circle of further creativity. A strong cultural sector and strong, sustainable digital industries, especially those powered by AI, go hand-in-hand. To this end, Vietnam has put in some important guardrails, such as requiring that any copyright protected works used for AI training (1) be lawfully accessed and (2) be made publicly accessible for the purposes of training of artificial intelligence systems (among other uses). Additionally, language has been adopted from the Berne Convention “three step test” that requires the use to “not unreasonably affect the legitimate rights and interests of the author”. So far so good, but these features may not be adequate to protect the interests of Vietnam’s creative industries unless made explicit. This is where careful drafting of the regulations comes into play.

First, on lawful access, this is positive in that it rules out the use of pirated content to train AI systems (if enforced) and should make it illegal to bypass a paywall (digital lock, or technological protection method, TPM) to access content. However, this provision provides only relatively thin protection since lawful access can be obtained by purchasing the cheapest or lowest minimal access possible and then using that access to justify commercial exploitation. For example, an AI company could purchase exactly one copy of each book that it wants to ingest, or one copy of a sound or audio-visual recording.

With respect to the requirement that content be made “publicly accessible” before it can be used for training, this implies that any content not made publicly accessible should be off limits to AI developers. Therefore, disclaimers or terms of service specifying what uses are acceptable must be respected along with technical indicators, such as robots.txt protocols, that mark content that is not to be copied. Again, regulation will need to specifically deal with these points to avoid any ambiguity. On the basis of the “trust but verify” principle, AI developers should be subject to maximum transparency requirements and must be required to document all protected content used for training.

To give effect to the language taken from the Berne Convention requiring the use to not unreasonably affect the legitimate rights and interests of the author, this could also be addressed by regulation to make the meaning precise. For example, AI outputs that substitute for the originals on which they were trained clearly prejudice the rights and interests of the rights-holder. It is not enough that an AI produced output does not directly mimic the original; if it displaces it in the market, it is also damaging the economic interests of the author.

Finally, the law indicates that the use of text and data subject to copyright “must also comply with the Government’s regulations.” It is far from clear what this means. The use of text and data already must comply with the terms of the legislation as described above. This phrase would hardly seem necessary unless it is intended to provide some sort of general override that would invalidate the guardrails embedded in the law. This should be clarified in regulation or dropped entirely.

Vietnam has a lively and thriving cultural sector that is a key manifestation of its cultural expression and sovereignty. The Vietnam Creator’s Coalition has estimated that the arts, entertainment and recreation sector employed over 281,000 people and contributed VND 55,694 billion (USD $2.12 billion) to the Vietnamese economy. The cornerstone of this economic strength and growth is Vietnam’s copyright framework. The National Target Program for Cultural Development has set a target for the cultural industries to contribute 7 percent of the country’s GDP by 2030. If copyright protection is inadvertently gutted by poor application of the new law, those targets will be unattainable and the sector, an important instrument of national expression, will suffer.

The other element of the content protection environment in Vietnam that continuously undermines both domestic and international rightsholders is piracy. Vietnam is a known centre of audiovisual piracy, hosting numerous sites that operate more or less openly. It has been called out by USTR’s Special 301 report on many occasions, most recently again in 2025. Initially the worst offender was a pirate operation going by the name of 123Movies. More recently it has been Phimmoi and FMovies. Enforcement is weak and when a pirate site is shut down, usually after prolonged effort by rightsholders, it is not long before it is back in business. It is not just international rightsholders who suffer. The mass copying of the film “Red Rain” (Mưa Đỏ), the 2025 Vietnamese epic historical war drama adapted from a novel about the Second Battle of Quảng Trị in 1972, a fierce and protracted battle in the Vietnam War, is Exhibit A for the damage wrought by domestic piracy, as outlined by the British legal firm Rouse. The inability or unwillingness of the Vietnamese authorities to take effective action against openly operating pirate sites is a blot on Vietnam’s IP record and undermines the credibility of the stated objectives of the new IP law.

From the overall thrust of the law, it seems that Vietnam’s lawmakers want to protect the country’s valuable cultural assets and creative energy while providing scope for AI developers to have access to protected content under clearly defined conditions that respect the legitimate rights of creators. The regulations that will be drafted over the next few months need to make this explicit and apparent or else the intent of the law could be subverted. Robust action is also required to end the free ride being afforded to content pirates. It is often said that the proof of the pudding lies in the eating. In this case, the proof of the legislation lies in the drafting of sound regulations to implement the new IP law effectively, while taking meaningful action to curb the scourge of AV piracy.

© Hugh Stephens, 2026. All Rights Reserved.

Deloitte’s AI Nightmare: Top Global Firm Caught Using AI-Fabricated Sources to Support its Policy Recommendations

Close-up of the Deloitte building sign, with greenery in the foreground and a clear sky in the background.

Image: Shutterstock

As we start a new year, 2026, it is a given that artificial intelligence (AI) is going to be the big issue for authors, publishers and the copyright industries generally. The issue of whether it is legal to use copyrighted materials to train AI platforms without the consent of rightsholders will continue to be fought out in courts and legislatures. The use of AI to create content will also continue as an ongoing issue, both the extent to which assistance from AI renders outputs non protectable by copyright and whether content produced using AI is reliable and trustworthy. Deloitte, by revenue the world’s largest consulting firm has just learned that lesson in spades. As well it should.

Would you hire an expensive consulting firm that used AI to supplement its research, didn’t inform you it was doing so, and when it got caught serving up AI-fabricated citations claimed that the false documentation in no way invalidated its policy recommendations and conclusions? I wouldn’t but apparently the Government of Canada has no such qualms, according to a Canadian Press story. We are talking about Deloitte, one of the world’s four largest consulting and accounting firms (the others being PwC, EY and KPMG), a company that charges a premium for its specialized services, and which ought to know better. It’s not as if Deloitte was caught just once with its hand firmly embedded in the AI cookie jar. First it happened in Australia, where the company was forced to reimburse the client, Australia’s Department of Employment and Workplace Relations, for a report that was reportedly “riddled with fake citations, phantom footnotes, and even a made-up quote from a Federal Court judgment.” It also pulled the same stunt in Canada where it produced a report on health care for the Newfoundland and Labrador provincial government. The 500-page report contained at least four citations of research papers that do not exist. These were used, with others, to support recommendations related to recruitment strategies, monetary incentives, virtual care, and the impact of the COVID-19 pandemic on healthcare workers. However, as the Independent, a local digital news outlet reported, Deloitte, having been caught redhanded, stated it is “revising the report to make a small number of citation corrections, which do not impact the report findings.”

What a joke! Of course the fabricated citations impact the report’s findings, as any first year university student knows. Frankly, this is a disgrace and I think Deloitte should be put in the penalty box for six months to a year as punishment, i.e. no government should contract with them until they learn to clean up their act. The irony is that Deloitte advertises itself as providing consulting services to governments to enable them to use AI effectively. Consulting firms are supposed to bring new expertise and perspectives to management problems that governments, in this era of cutbacks, no longer have the resources to solve. They are expensive but provide a quick turnaround for public service managers who don’t have the in-house resources to deal with emerging issues, and who often don’t have the luxury of time to staff up to meet immediate needs. But the dirty little secret is that in many cases the consulting firms apply their cookie-cutter templates to inform their findings whether the template suits or not. They also employ junior staff to do much of the grunt work without, apparently, providing them with adequate supervision or guidance. But even if the labour-intensive task of finding citations to justify the “researched conclusions” of the commissioned report was subcontracted to an AI bot, someone senior at Deloitte signed off on the final product. It didn’t take the client or journalists very long to track down the fabricated citations, so why couldn’t Deloitte have run the same quality check? Because they couldn’t be bothered, I guess.

Despite having been caught, Deloitte may be big enough to shrug this one off, but I sincerely hope they have learned a lesson. Even one false or fabricated citation undermines the credibility of research. As I noted in a recent blog post (Delegating Research to AI is a Risky Proposition: The “Hallucination” Phenomenon-User Beware), “In our rush to embrace AI, many seem to have forgotten the value of human creativity and judgement”. Deloitte has egg on its face, and needs to wear this. A consulting firm is only as good as its reputation, and as far as I am concerned, Deloitte has just put its reputation through the shredder.

This is a cautionary tale, one that has enmeshed not only the world’s largest consulting firm, but various law firms that have been caught citing fabricated precedents. Students would be sanctioned for using AI this way (if they were caught) and academics would suffer major hits to their reputation. Research results and qualifications might be invalidated. If these are the sanctions for misuse of AI, then we should expect no less from entities like Deloitte and its ilk. Let’s hope there are no further AI fabrication horror stories in 2026. (A vain hope, I am sure).

© Hugh Stephens, 2026. All Rights Reserved

The Year That Was: Looking Back at Copyright Issues in 2025

A graphic featuring the text '2025 A Look Back' inside a magnifying glass on an orange background.

Image: Shutterstock.com

In last year’s annual retrospective, I commented that just about the only significant copyright story in town was the impact of AI on copyrighted content. There were three primary dimensions, unauthorized use of copyrighted content for AI training, copyright infringing outputs produced by AI platforms and whether works produced by or with AI qualify for copyright protection. The first two elements, especially the unauthorized use of copyrighted content for AI training, continues to be the big story in 2025. Little has changed in terms of the fundamental issues, although this year there were two key court rulings in the US that provided some legal guidance. These were the Bartz v Anthropic and Kadrey et al v META cases, both decided within hours in the same courthouse in San Francisco, but by different judges. As I noted in a blog post at the time, in July, the results were a very mixed bag. (Hold the Champagne: The Two AI Training/Copyright Decisions Released in the US Last Week Were a Mixed Bag for AI Developers). While both judges found the copying of copyrighted works to train AI algorithms “transformational”, and thus tending toward fair use, in the Anthropic case the AI developer was castigated for initially using two pirate data bases (Library Genesis, aka LibGen, and Pirate Library Mirror, aka PiLiMi), as source material. This use threatened to result in hundreds of millions of dollars in statutory damages for authors whose works were included in these databases (if they had registered their works with the US Copyright Office). Thus, it was with no great surprise that the world learned in September that Anthropic had agreed to a $1.5 billion settlement to bring an end to the case. (When the End Does Not Justify the Means: Anthropic’s $1.5 Billion Lesson). In the Kadrey case, the judge suggested the plaintiffs should have made a stronger market dilution argument, which might have overridden the transformation finding. (New AI works produced based on unauthorized inputs of copyrighted works will devalue and dilute the market for the original works). That theory has yet to be fully tested.

While the Anthropic settlement seemed large in absolute terms, many criticized it as just another cost of doing business that failed to protect authors. But that is what licensing is, a cost of doing business. The damages could have been larger than the settlement, so Anthropic made a strategic decision to settle. Given the precedent, META must be worried given the clear evidence that its top executives gave the green light to use LibGen for training purposes despite internal warnings about the risk of doing so. (Is it Ethical to Use Pirated Content for Commercial Purposes? META Thinks So). The Anthropic decision gave impetus to ongoing licensing discussions between AI developers and content owners, especially large corporate rightsholders, like news media, music labels and big studios, such as Disney. The list of companies striking significant licensing deals with AI developers, either for AI training or outputs containing derivative content, is rapidly growing. Meanwhile lawsuits, such as New York Times v OpenAI, continue in lieu of settlement agreements.

Outside the US where the “fair use” doctrine along with its transformation interpretation doesn’t apply, there are also lawsuits against AI developers (in Canada and India for example). However, in these “fair dealing” countries AI developers are in a potentially tighter spot. In many countries there is no statutory exception to allow unauthorized access to copyrighted content for AI training, commonly known as text and data mining (TDM). Where there is such an exception (as in the UK or EU), it is confined to very specific circumstances such as non-commercial research, or else it requires that rightsholders be provided with an opt-out mechanism. Given these inconvenient facts (and their inability to rely on fair use arguments outside the US), AI developers have been mounting a full-court press to have TDM exceptions introduced into national legislation using the pretexts that a) everyone is doing it (which is certainly not true) and b) if governments don’t create an exception allowing TDM for AI training, all those AI development funds will flow elsewhere. In other words, throw rightsholders and creative industries under the bus in order to chase some ephemeral AI research funds. Australia has just rejected this binary approach, announcing that a TDM exception will not be part of its review of copyright laws to help address the needs of the AI industry. A number of Asian countries are reviewing the need for a TDM exception, but are rightly being very cautious not wanting to sacrifice the vital cultural and economic interests their creative sectors represent. Canada is another country without a TDM exception, a situation that has made OpenAI jittery since they are being sued in Ontario Superior Court by a consortium of Canadian news publishers for copyright infringement, bypassing Technical Protection Measures (protected paywalls) and breach of contract. OpenAI has been trying to get the case moved to the US by challenging the jurisdiction of the Canadian court, so far unsuccessfully.

Just as lawsuits in the US are providing the impetus for settlement discussions between AI developers and rightsholders, lawsuits against the AI industry outside the US will potentially have the same effect. The big story of 2025 is how many licensing agreements have been already reached. Even META, which in Canada and more recently Australia insisted it doesn’t need news content and would not pay for it, has reached a news media licensing deal with a number of companies including USA Today, People, CNN, Fox News, The Daily Caller, Washington Examiner and Le Monde. Back in 2021 META agreed to license news content in Australia as a result of the introduction of the Australian News Media Bargaining Code, but when Canada enacted a similar provision, it refused to do so and evaded the obligation to pay for local news by blocking it on its Facebook and Instagram platforms. Subsequently, when the Australian Code came around for renewal this year, META balked. Australia is still contemplating its next steps and while it is treading carefully, indications are that it is prepared to move against META.

One reason why Australia is moving carefully is the Trump factor. Trump’s erratic behaviour including the tearing up of the established rules of international trade is the second big copyright theme of 2025, after AI’s predations and encroachments. Not that Donald Trump knows much about copyright or understands it, but in his broadsides against trading partners and the international trading system, copyright industries inevitably become caught up in his web. Whether it is threatening tariffs on movies filmed outside the US, regardless of the fact that the largest slice of the box office comes from non-US sources and many films are either co-financed by offshore producers or require non-US settings (Trump’s Threatened Tariffs on Hollywood Films Produced Outside the US: The Medicine Could be Worse than the Disease), or taking aim at the policies of other countries that may have a digital or content component, like the Digital Services Tax in Canada,and its Online News Act, Trump is unpredictable and often off-target. His initial tariff measures against Canada included everything from steel to autos, based on the specious “national security” argument that there was a flood of fentanyl coming into the US from north of the border (actually about 0.2% of the total in 2024), but so far he has not specifically targetted the cultural industries. That may all change as the negotiations for renewal of the CUSMA/USMCA begin in the new year and as Canada struggles to redefine “Canadian content”.

There are industry groups in the US that claim provisions of the Online Streaming Act violate the terms of the CUSMA/USMCA by discriminating against US content providers, and that the only way Canada can justify these provisions is to invoke the cultural exception clause of the CUSMA. This would allow the US to retaliate in any sector with equivalent commercial effect. I have questioned this interpretation but we will probably never know who is right because the Online Streaming measures are unlikely to be challenged by the US prior to the start of CUSMA/USMCA renegotiations in early 2026. In those negotiations, the US will almost certainly take aim at various elements of cultural legislation like Online News and Online Streaming, along with other issues like dairy supply management. These may or may not be subject to negotiation as part of the renewal of CUSMA, if indeed the US or Canada are prepared to renew it.

Even if Canadian actions are consistent with the terms of the Agreement, this is essentially meaningless given the way the Trump Administration operates, as demonstrated earlier this year when Trump overrode various tariff free commitments in the Agreement on the basis of unilaterally declared national security concerns. These included the specious fentanyl trafficking claim mentioned above. Not only did the miniscule numbers give the lie to this assertion but the US claim somehow made Canada responsible for ensuring the security of the US border. As professor Fen Hampson of Carleton University has eloquently pointed out, the fate of CUSMA/USMCA depends not so much on where the economic interests of the United States lie, but on the fickle and unpredictable whims of Donald Trump personally.

As we head into 2026, there will be more Donald Trump uncertainty affecting copyright industries, more AI disruption, more lawsuits, more settlements and more licensing agreements to avoid lawsuits and settlements. That’s a pretty safe prediction based on looking back on 2025.

© Hugh Stephens, 2025. All Rights Reserved.

Flash Fiction at Christmas: The Bench

A wooden bench on a rocky surface surrounded by greenery and sparse trees, with a cloudy sky in the background.

Photo: Author

In just a couple of days, gifts that have been gathering under Christmas trees will be opened. Among them will be books of various genres; children’s books, novels, cookbooks, historical biographies, poetry and so on. While I doubt there will be many books of “flash fiction”, I am today offering a flash fiction Christmas present for everyone. Read on.

Well, you may say, that is all well and good, but what exactly is flash fiction? There are many ways for aspiring writers to take the plunge into getting published; sketching out a poem, working on a short story, tackling a novel. All have their challenges. Somewhere in the middle is “flash fiction”, a genre that has become increasingly popular in the digital age (when the ability to concentrate on any content seems to have diminished exponentially as more short-form digital content becomes available; tweets–remember them?–, Youtube reels, news snippets, AI summaries, and so on). Flash fiction resides somewhere between the short story and the microstory, with varying definitions. One I have seen limits the genre to 1500 words or less; other definitions place the limit at 300 words. My first engagement with flash fiction came recently when the Victoria Writers Society announced a contest for flash fiction entries not exceeding 500 words. I thought I would give it a try. I had joined the Society because they had been generous enough to let me speak to them about copyright, based on my book, In Defence of Copyright.

I was intrigued because I have never written fiction. While I routinely crank out a weekly blog (usually between 1200-1800 words, often struggling to keep it shorter rather than longer) I have always admired those who work for years on their opus. Writing a novel is no doubt a struggle unless you have the formula to churn them out like the late romance writer Barbara Courtland, who is reported to have produced over 600 titles that sold 650 million copies in various languages. She could produce a book in two weeks, always tied to the same formula of the impoverished heroine who meets her Prince Charming, marries (what did you expect?) and lives happily ever after. Of course, today AI can outdo that record, but I doubt if AI slop will ever be able to best Dame Courtland’s success.

Daunted by the task of attempting a novel, or even a short story, (and being inherently lazy), I thought that taking a run at creative writing within the limit of 500 words would be a good place to start. It wasn’t that easy. As Mark Twain is famously reported to have said, “I didn’t have time to write a short letter, so instead I wrote a long one.” Getting a story crammed into 500 words is a challenge. Apart from structuring the plot and ensuring a suitably arresting denouement, every word counts. It is a really good exercise to draft a piece and then go back and excise the flab. I found it amazing how much redundancy can be chopped. Editors already know this and are masters of making works more concise. Feedback was provided reminding entrants that;

“Each story still needs a clear conflict and a clear narrative arc, where the characters are changed or transformed in some significant way by the last sentence. An event simply happening or characters having an experience does not offer enough narrative movement to push a story along.”

After initially just thanking me for my efforts, to my great surprise the VWS then selected my entry, along with several others, to be published in their journal, Island Writer. You can judge for yourself whether the selection committee made a wise choice. While you could read my opus by purchasing the publication, for those readers outside the tight little circle of southern Vancouver Island, here it is. A Christmas present for you.

The Bench

“The chip path was soft underfoot. Ochre rays of the setting sun illuminated the salal and cedar ahead. A turn in the path and he emerged onto a headland where a gust of wind brought just a whiff of salt and seaweed. Across the water he could see the sun catch the snowcapped peaks of the Olympic range, tinting them pink. A large container ship, its lights sparkling, was highlighted against the darkening sea. Ahead he saw a viewing bench looking out across the water. At first, he thought there was no-one there, but then he saw a silhouette and realized someone was seated, facing the water, apparently reading a book. It was a woman.

He thought of the many times over the years he had walked this very path and inhaled this view. It had always revived and refreshed him, at times of stress or at times of reflection. He thought back on the different events in his life that had brought him to this place of repose. University studies, and the uncertainty that accompanies the future. Which career path? Will there be a partner to share the journey? Walking hand in hand with her, breathing in a whiff of fragrance from her hair. Putting their boy on his knee and pointing out to sea, looking at the freighters and sometimes a sailboat. Strolling in the evening with her before putting him to bed. Family picnics. Sitting on that very bench with the boy, now a teenager, listening to his growing pains, trying to offer counsel, building hope for the future. Celebrating his graduation with photos against a sea and mountain backdrop. Three generational family gatherings. And times of sadness too. Now once again, on this early April evening, he was alone. This was a place of a lifetime of memories.

The bench was just up ahead. He thought the woman was still there, but when he reached it as the shadows lengthened, she was gone. He knew this bench and its inscription. It had been there for years. Like most of the benches in the park, it had been donated in memory of someone who had once loved this spot. It said, “In loving memory of Lara Porter, April 1930-September 1995. This place brought her comfort”. Then he noticed the small book lying on the bench. It was a collection of poems by Dylan Thomas. It lay open at the work “Do Not Go Gentle Into That Goodnight”. He picked it up and read;

Do not go gentle into that good night

Old age should burn and rave at close of day

Rage, rage against the dying of the light

Flipping to the front of the book, he noticed the inscription. “To Lara, Much Love on your birthday, April 6, 1995”. Then the sudden realization. Today was the 6th of April. At that instant, the sun slipped beneath the horizon, continuing its journey west. He turned to see a radiant glow.”

There it is. The setting is Saxe Point Park in Esquimalt, overlooking the Strait of Juan de Fuca and looking across the water at the Olympic Range in Washington State. The story is a total fabrication. I did not grow up in Victoria and did not have a son. But there is a bench there inscribed in someone’s memory, but not Lara Porter.

It was an interesting exercise. Will I move on to a short story? I’m not sure. As I said, I am inherently lazy. But if you want to try your hand at writing, try a 500-word flash fiction. It’s a challenge, but fun. Go for it. Merry Christmas.

© Hugh Stephens, 2025. All Rights Reserved.

No Surprise:  Ontario Court Asserts Jurisdiction in Canadian Media Lawsuit Against OpenAI

A judge sitting at a bench in a courtroom, wearing a black robe with a red collar, Canadian flags in the background.

Image: Shutterstock

The Ontario Superior Court has ruled it has jurisdiction to hear the case against ChatGPT owner OpenAI brought by a consortium of Canadian media companies led by the Toronto Star. The media enterprises, who include the Globe and Mail, PostMedia, CBC/Radio Canada, Canadian Press and Metroland Media Group, are suing the US company for copyright infringement, circumvention of technological protection measures (TPMs), breach of contract, and unjust enrichment as a result of OpenAI’s scraping of their websites to obtain content to train its AI algorithm. The allegations also cover OpenAI’s use of Retrieval Augmented Generation (RAG) to produce contemporary search results from paywall-protected content that augment ChatGPT’s AI-generated responses. When the suit was brought in November 2024, OpenAI had challenged the Ontario court’s jurisdiction on the basis, among others, that it had no physical presence in Canada. As pointed out by this legal blog, a court may presumptively assume jurisdiction over a dispute where one of five factors is present:

  • The defendant is domiciled or resident in the province.
  • The defendant carries on business in the province.
  • The tort was committed in the province.
  • A contract connected with the dispute was made in the province.
  • Property related to the asserted claims is located in the province.

The court found that OpenAI carries on business in Ontario notwithstanding its lack of a physical presence and was a party to contracts in Ontario as a result of tacitly accepting the terms of service regarding access to the media companies websites when it scraped them.

OpenAI wanted the venue of the litigation changed to the United States to take advantage of developments in US law regarding unauthorized reproduction of copyright protected content for use as AI training inputs. To date, while many cases are still ongoing, US courts have tended to support a fair use argument by AI developers allowing them to access copyrighted content without permission on the basis that the end use is “transformational”, resulting in a new product that does not compete with the original work. In Canada, the fair use doctrine does not apply and exceptions to copyright protection are either explicitly laid out in the law (e.g. for law enforcement or archival preservation purposes) or are governed by the fair dealing provisions of the Copyright Act. These require that an unauthorized use fall into one of eight categories (research, private study, education, parody, satire, criticism, review and news reporting) that is in turn subject to various court-interpreted criteria such as amount of the work copied, the purpose of the copying, market impact etc. AI developers have been lobbying for the introduction of a text and data mining (TDM) exception into Canadian copyright law, but so far this has been successfully resisted by Canada’s creative community. All this to say that it is more difficult for AI companies to avoid liability for unauthorized use of copyright protected material in Canada than in the US, thus the importance of whether the Ontario court has jurisdiction.

Back in September, on the basis of previous Canadian court rulings where courts ranging from provincial courts to the Supreme Court of Canada asserted jurisdiction over large digital US companies operating virtually in Canada, such as Google (who challenged Canadian legal authority over them on the basis of lack of a physical presence), I predicted (guessed would be a more accurate term) that the Ontario court would be loath to surrender jurisdiction simply because the company was headquartered in the US. The earlier cases were for defamation rather than copyright infringement, and my “prediction” was based more on a hunch than legal analysis, but I am satisfied that I called it right. OpenAI has no compunction about selling services and collecting revenues in Canada and presumably (I hope) pays taxes here, although it is not subject to the Digital Services Tax (DST) that the Carney government threw overboard in a vain attempt to placate Donald Trump. Recall that Trump had threatened to terminate trade talks if Canada proceeded to implement the long-planned DST, so Canada blinked. Trade talks resumed until Trump found another excuse to end the talks, in this case the anti-tariff ads on US television placed and paid for by the Ontario government to which he took offence. But there is no doubt that OpenAI does business here; it just doesn’t want to be subject to Canadian law and Canadian courts. It can’t have it both ways.

While this is a victory for Canadian sovereignty, just because the Ontario Superior Court has confirmed its jurisdiction, this doesn’t mean that once the substantive proceedings begin copyright infringement will be found. Lawyer Barry Sookman, in an analytical  blog post on this topic, has noted that in determining whether the alleged copyright infringements occurred in Canada, “the court relied heavily on the Supreme Court decision in SOCAN for the proposition that the territorial jurisdiction of the CCA (Canadian Copyright Act) extended to where Canada is the country of transmission or reception.” However, “SOCAN applied the real and substantial connection test to the communication to the public right” whereas the alleged copying involved the right of reproduction.

Sookman continues;

“…that test does not apply to the reproduction right. (The Federal Court has) held that the only relevant factor is the location in which copies of a work are fixed into some material form. The locations where source copies reside or acts of copying onto servers located outside of Canada, are not infringements” (according to the cases cited).

Inside baseball information but important when it comes to determining copyright infringement. On the other hand, it seems to me that the infringement involved not just, potentially, the reproduction right (the copying) but also the communication right, because OpenAI, through Microsoft, provided RAG content to users in Canada and elsewhere purloined from behind the paywalls of the media companies. So, we will have to see. Lots of fodder for IP lawyers.

In the meantime, deep-pocketed OpenAI will appeal the jurisdictional ruling—and will likely lose again. The appeal will buy time for it to negotiate licensing deals with the complainants. This is increasingly the model in the US as AI developers, including OpenAI, are reaching licensing agreements with content owners, particularly media organizations. To date, OpenAI has signed licensing deals with the Associated Press, the Atlantic, Financial Times, News Corp, Vox Media, Business Insider, People, and Better Homes & Gardens, among others, while being sued (in addition to Toronto Star et al), by the New York Times and a collection of daily newspapers consisting of the New York Daily News, the Chicago Tribune, the Orlando Sentinel, the Sun Sentinel of Florida, San Jose Mercury News, The Denver Post, the Orange County Register and the St. Paul Pioneer Press. Even META, that arch-opponent of paying for media content–which it claims adds no value to its users– has struck a media deal with news publishers, including USA Today, People, CNN, Fox News, The Daily Caller, Washington Examiner and Le Monde. (One wonders if this will cause it to rethink its position of thumbing its nose at Canada’s Online News Act, where it “complied” with the legislation by blocking all Canadian news links).

In another content area, OpenAI and Disney have just agreed on a three-year output deal, allowing it to use Disney characters (subject to certain limitations) in its AI creations. (Meanwhile Disney is suing Google for using its characters in Google’s AI offering). Open AI is currently facing 20 lawsuits, including the Toronto Star case, and needs to resolve these legal challenges before its expected public offering next year or 2027. The spectre of impending lawsuits will inevitably lower the IPO price.

Most if not all of these lawsuits are going to end in settlements via voluntary licensing agreements, but that will only happen if OpenAI thinks the alternative (losing a major lawsuit) is a worse outcome. If it can wriggle out from the Toronto Star case by invoking some specious argument related to jurisdiction, it will. If it can’t it, will eventually open its chequebook and provide the Canadian media outlets some compensation for the valuable curated content it has hijacked. Canadian courts need to stay the course to help ensure that this happens.

© Hugh Stephens, 2025. All Rights Reserved.

Delegating Research to AI is a Risky Proposition: The “Hallucination” Phenomenon (User Beware)

A graphic showing a cartoon robot head on a computer monitor with the text 'The World is Flat Because I Say So' in a playful font.

Image: Shutterstock (author modified)

It seems everyday new applications and new threats emerge from the AI world. This applies in particular to creators who see growing AI challenges to their livelihoods; graphic art and album covers spat out by AI generators; voice actors replaced by AI clones; authors struggling to make their works known in a sea of AI-generated slop; now AI artists are even making the Billboard charts. At the same time, AI has many other functions and produces a host of products that have little to do with artistic creation. In particular, it can be used as a crutch to assist and enable research in a wide range of fields. Today it is routinely used by everyone from school kids to law firms to health care researchers. And that is where the risks of mainlining AI are the most evident, because of the propensity of AI platforms to fabricate plausible sounding misinformation.

In a blog post earlier this year (AI’s Habit of Information Fabrication (“Hallucination”): Where’s the Human Factor?) I discussed some examples of law firms caught submitting non-existent case precedents in court as a result of sloppy legal research using AI. Judges have very limited tolerance for this practice, which wastes valuable court time, and they are increasingly imposing significant penalties—that is, if the fabricated information is actually spotted. The problem is not going away. This website maintained by Paris-based legal scholar Damien Charlotin has compiled a database of more than 550 legal cases in 25 countries where generative AI has produced hallucinated content.  These are typically fake citations, but also include other types of AI-generated arguments. The US wins the lottery at 373 cases, but Canada is second with 39. Even Papua-New Guinea has one case.

As you can well imagine, AI hallucinated results in health care could be fatal. As Dr. Peter Bonis, Chief Medical Officer at Wolters Kluwer Health points out, hallucination in the health care field has led to various consequences such as recommending surgery when it was not needed, advising that a specific drug could be safely stopped abruptly when this was known to be dangerous, avoiding recommending vaccinations based on known allergies even though it was safe to do so, proposing wrong starting treatments for patients with rheumatoid arthritis and so on. You get the picture. You don’t want your family doc using AI search for the remedy for whatever ails you. The fact that the models present incorrect information with such confidence, and that potentially dangerous incorrect information is embedded with a lot of correct information, makes proper use of AI outputs particularly challenging.

How is it that AI platforms consistently produce unreliable results? This MIT Sloan article identifies three elements;

  • Training data sources (the uneven quality of inputs, including pirated, biased and otherwise unreliable content)
  • Limitations of generative models (generative AI models are designed to predict the next word or sequence based on observed patterns and to generate plausible content, not to verify its accuracy)
  • Inherent Challenges in AI Design (The technology isn’t designed to differentiate between what’s true and what’s not true)

This is all pretty concerning if people are going to surrender personal judgement to AI and use it to cut corners without verification. One way to address part of the problem is to ensure the training data used is reliable and of high quality. That is where licensing of accurate, curated data and content as training inputs is important and that is why a licensing market is developing as AI companies seek out better quality data to distinguish their product from that of their competitors. This can be very helpful where the AI platform is limited to discrete areas of knowledge, such as in the medical field for example, where usage can be limited to professionals who are prepared to pay for a bespoke AI product and who are qualified to interpret the results properly. AI for the general public is another matter, and this is where most of the problems arise. Unfortunately, while improving the quality of training data helps reduce hallucinations, it does not completely eliminate them. As the New York Times has reported,

“Because the internet is filled with untruthful information, the technology learns to repeat the same untruths. And sometimes the chatbots make things up. They produce new text, combining billions of patterns in unexpected ways. This means even if they learned solely from text that is accurate, they may still generate something that is not.”

User beware. Nonetheless, better inputs lead to better outputs. As AI developers work to take their products to the next level by refining their training processes and making outputs more predictable and trustworthy, they will need access to curated, proprietorial content and closer collaboration with content owners. Dr. Bonis noted that for specialized areas like health care, AI companies will get better quality feedstock while creators of the content will receive funding allowing them to continue research. A virtuous circle.

Users bear a big responsibility to ensure AI is employed effectively. The mindless, unjudgemental use of AI to reach conclusions in areas where the user has little knowledge can be dangerous. By all means use AI as a tool to sort and categorize, but don’t rely on it to produce the answers on which substantive decisions will be based. Any sensible user of AI has a pretty good idea of the answer to the question before it is even asked. It is also a good idea to refine the question, so you narrow the range of possibilities.

Some proprietary AI models offer RAG (Retrieval Augmented Generation) where the AI will retrieve relevant information from trusted sources to supplement its preliminary analysis. This can increase reliability. However RAG, where the AI goes after specific inputs to bolster its results, can also expose AI developers to charges of copyright infringement, as is currently the case with Canadian AI company, Cohere, which is being sued by a number of newspaper publishers, including the Toronto Star, for copyright infringement. As Canadian lawyer Barry Sookman has  pointed out in a recent blog, use of RAG can create risk for the AI platform. In the case of Cohere, when its RAG feature was switched on, it reproduced large amounts of almost verbatim text pulled directly from the litigating news sources. But if the RAG function was switched off, it produced fabricated information (hallucinations) yet still identified this false information as coming from an identified reliable news source, leading to charges of trademark dilution. The value of the brand was diminished by the attribution of false information to it. This trademark dilution issue is also part of the New York Times case against OpenAI.

At the end of the day, it is a case of user beware as a recent case in Newfoundland demonstrates well. The Government of Newfoundland commissioned an in-depth study on the future of education in the province. The 410 page report containing over 110 recommendations, authored by two university professors, was released with great fanfare at the end of August. No doubt a great deal of careful research had gone into producing the study over the 18-month production period. But then cracks started appearing in the edifice. It was chock full of made-up citations. The more people started checking, the more they found. The Department of Education and Early Childhood Development tried to whitewash the issue by saying it was aware of a “small number of potential errors in citations” in the report. But even one fabricated citation is one too many! If you search for the report online now you get the classic “404 Not Found” message. A lot of work has potentially gone down the drain, and possibly the credibility of two academics has been destroyed by careless use of AI. This is a cautionary tale that I have no doubt will be repeated.

In fact, it was repeated just a few days later. It seems Newfoundland is particularly prone to victimization by hallucinating AI platforms. After the education report debacle, new reports have surfaced that a $1.5 million study on the health care system conducted by none other than Deloitte also contains fabricated information included made up references. The opposition party is demanding the government insist on a refund.

In our rush to embrace AI, many seem to have forgotten the value of human creativity and judgement. Coming back to the creative industries and AI, some of those whose livelihoods may be threatened by this new phenomenon are bravely trying to find a silver lining. Some voice actors are generating an additional revenue stream by licensing their voice clips for AI training, and many graphic artists use AI as an assist. Are they putting themselves out of work in the long run or are they simply adapting? The jury is still out, but the generally low quality of AI produced art, music and literature, as well as the ongoing problem of hallucination, suggests that there will always be a need for real human input. Anyone planning on substituting AI for “real work” had better think again.

© Hugh Stephens, 2025. All Rights Reserved.