The Budapest Open Access Initiative is more than 25 years old. It called for journal content to be “freely accessible online…without expectation of payment.” It also specified that articles be free for use “as data to software.” Publishers moved to adopt OA business models — including defaulting to CC BY licenses — at a time when text- and data-mining (TDM) and AI were not generally part of the conversation. With AI at the fore, some publishers have ceased defaulting to CC BY. This has been driven both by author skepticism and fears of lost revenue.
In 2008, the NIH began its mandate that accepted articles be deposited into PubMed Central and made available 12 months after publication of the Version of Record. Many large publishers agreed to deposit articles on behalf of authors based on the verbal agreement that the NIH would not subsequently shorten the embargo period. None of those involved in this original understanding are still involved, and US federal grants have moved to zero-embargo green mandates. This has been coupled with a proposal by the US government to cease funding publishing from grants altogether. Different versions of this situation are under consideration elsewhere. Again, publishers, authors and institutions are rethinking their policies.
Gold APCs and transformative agreements, once the darlings of the OA movement (and certainly a driver of OA adoption), have been criticized for everything from failing to harm big publishers to limiting authorship from the Global South. Models like Subscribe To Open, created in partial response, have both succeeded and failed.
So, today we ask the TSK Chefs: As participants in scholarly publishing, and with the benefit of 20/20 hindsight, what went right, and what could have/should have been done differently?
Roy Kaufman
The following represents my personal views and not those of my employer.
One of the fundamental truisms of post hunter/gatherer society is that things cost money. Money, which is an accepted unit of stored value, is distributed unequally. This too seems hard to debate. I would go further to state that the scientific publishing community writ large does not have the power or authority to alter the truth of the prior statements.
Why do I start there? Because at the beginning of the OA movement, some adherents spoke in terms of ethical belief systems. If you were not actively trying to move everything to the then-popular fully OA model, you were somehow morally culpable — regardless of the economic consequences.
We still see shades of this in the positioning of specific OA business models. For example, subscribe to open (S2O) is a model that has been branded “equitable OA,” as if to say that if you are not employing S2O or a model with similar attributes, you are against equity. S2O is a great model for some journals — especially as it offers a fast route to OA. It does not require approbation or guilt as a justification. As we should now know, each model shifts costs, risks, and rewards, but none eliminates them. A recent article in Nature discussing Diamond OA, which itself is often posed as the “most ethical” model, emphasizes this.
In Budapest, idealists and opportunists met and agreed on a definition of OA that included broad reuse rights, including machine reuse. This started wealth transfer from the publishing community (publishers, authors, libraries) to, among others, tech companies. Budapest was supported initially by publishers who had little (or no) third-party rights businesses. As such, they could readily support CC BY licensing without harm and maybe even receive a competitive advantage over established publishers. This does not mean they did not believe in it; merely that it was a relatively easy stance.
Making content available on preprint servers (green OA) contributed to the wealth transfer. While preprint servers provide genuine benefits for the early sharing of knowledge, they also allow indiscriminate reuse of content (e.g., AI), become full of AI-generated papers, and provide a way for bad actors to disseminate misinformation.
Open access is great when it is sustainable. Gold, Diamond, and S2O all have pros and cons. OA has enabled publication growth. In the absence of OA models, the added costs of publication growth would have been passed onto an already taxed subscription system — which might not have been able to bear it. However, we need to acknowledge that while some journals benefit from S2O, others will never move from hybrid status. Variation and choice are key to a robust system whose ultimate goal should be to advance human understanding and knowledge, not to support one business model over another.
What did we get right? We provided new models which gave authors, institutions, and funders more choices to help the publication system grow with the growth of research. What did we get wrong? We pretended that all disciplines, journals, and business models could be crammed into one model — and one license — to rule them all. We then wrapped it in moral terms that unnecessarily limited debate. I’d like to think we are moving past this.
Haseeb Irfanullah
The “open access” movement, as we currently define it, has generated some amazing numbers over recent years (e.g., Gold OA percentage increased from 17% (2015) to 44% (2025)). But, to me, it has failed because we wanted to make something “open” to all, without addressing the underlying structural reasons that “closed” it in the first place.
Scholarly publishing is an outstandingly profitable business where publishers make money by adding value to the submitted manuscripts. (These manuscripts magically appear on publishers’ portals or editors’ tables, because the academic world continuously insists on its researchers in doing so). Publishers’ value addition takes place by maintaining infrastructure for manuscript submission to paper publication; by (literally) arranging ‘blind dates’ between the peer reviewers and the authors; providing all sorts of cosmetic (editorial) inputs (massive alterations are always done by the authors); and constructing the so-called prestige of a journal/publishers which essentially determines if an open access journal can get away with either charging US$ 200 or US$ 12,000 per article as article processing charges (APCs).
Therefore, when the open access campaign wanted to democratize access to research, access inevitably increased, but publishers did not sacrifice a single cent, fen (in China), paisa (in South Asia), or penny. Whatever open access modality we talk about — authors paying APCs, institutions paying publishers under transformative agreements, hybrid journals accommodating both poor and rich authors in the same journal issue, or libraries’ subscription fees being used to make journals open (S2O) — publishers’ profit margins remains amazingly unstressed.
We wanted to drastically change a system from closed access to open access without financially paying significantly for that shift. But as a result, we have been paying the price with the surge of controversial megajournals, the experiments with APCs and other models, the so-called predatory journals, the rampage of papermills, the money-making special issues, the record growth of retractions, and the delisted or suspended journals.
I believe our basic perception of making academic journals accessible to every person in the world in the name of open access is fundamentally flawed. Do we have any data on how many common, non-academic people read open access articles, and how they use the new knowledge recorded in them? Let’s admit, open access or not, the almost exclusive use of journal articles is to be cited by other journal articles in the future. That’s why almost all metrics measuring journals’, even researchers’, reputation are solely based on citation numbers. In the same vein, publishers and their associations do not appreciate the millions of pieces of grey literature and preprints in open access conversations, because these are not peer-reviewed. The Global South publishes hundreds of free access journals full of valuable national and regional knowledge, but we undermine them not being indexed by reputed indexing agencies. Further, the whole publishing system is not only flawed, but is maintained by false premises. On the one hand, we call knowledge a common good, but tie it either to subscriptions or APCs. We campaign for open access, but can do nothing when publishers make this movement a money-making, rebranding opportunity. Journals publish new knowledge and evidence on numerous issues to make society better, but the publishing industry is continuously overlooking its own exploitative, excessive, profit-driven, unjust practices.
Therefore, the limitations and failures of open access cannot be discussed in isolation, ignoring the whole publishing industry’s moral poverty. The question is not what we could do differently to make our open access effort better; the question is more basic. The current size and article production rate of journal publishing is unreal. We desperately need a disruption and to start a “planned degrowth” of academic publishing by focusing on quality over quantity. This would replace the ‘publish or perish’ mindset with ‘publish quality papers to flourish’. Recruitment and promotion rules of institutions should consider an applicant’s quality of research, research outreach, and impact on the ground, rather than the number of papers and their impact factors. These may potentially lead to a new level of challenges in accessing global research. We have to face those challenges if we want to grow out of the situation where our pieces of research are fed as numbers in some metrics. We must break this numbers game.
Hong Zhou
I wouldn’t have made research less open. I would have been more ambitious about what openness enables — and more deliberate about how we sustain it.
Making research available beyond subscription barriers was worth doing. In my previous post, Approaching AI and Open Science in Sync, I argued that AI and open research should develop together. With hindsight, I would have translated that thinking into publishing requirements and business models much earlier.
First, I would have made data, code, software, notes, procedures, and methodologies first-class, mandatory components of research contributions wherever applicable, with justified exceptions and protections for sensitive materials. I would capture them during research and authoring, rather than reconstruct them at submission, and make them citable, versioned, and preserved.
I would invest as seriously in verification as in publication efficiency. My goal would be to enable researchers, reviewers, and AI agents to inspect evidence, securely execute analyses, and compare outputs with reported results — starting before formal peer review. I would support independent replication too, without pretending that rerunning the original analysis amounts to a new experiment.
Second, I would ask for structured knowledge alongside the narrative manuscript: concepts, events, relationships, and claims linked to evidence, methods, uncertainty, and limitations. These author-validated contributions could accumulate into interconnected, domain-specific knowledge graphs, supporting human understanding and AI retrieval and reasoning. I would design them to preserve disagreements and incorporate corrections — not present every published claim as settled fact.
Authors should not have to become knowledge engineers. Integrated tools and AI-assisted extraction should make this practical, with researchers checking meaning rather than manually encoding their papers.
Third, I would design a broader mix of funding, licensing, and services, rather than treating APCs or transformative agreements as the destination. I would anticipate not only reading, but AI model training and fine-tuning, automated discovery, and paid knowledge services. CC BY already permits commercial reuse under its terms; it is not simply “free to read.”
For me, openness and commercial value need not be opposites. The opportunity is to distinguish what is openly reusable, what requires additional permission, and what customers pay to have done for them. Curated training datasets, maintained knowledge graphs, and reproducibility services could provide genuine additional value. I would build AI gateways around these capabilities, with clear permissions, attribution, and transparent arrangements explaining how authors’ interests are protected, and revenues support research stewardship.
Finally, I would judge progress by participation, not just OA percentages. Moving a barrier from readers to authors — or to expensive AI tools — is not the outcome I would want. Shared infrastructure, affordable services, and recognition for data, software, and methods would be part of the plan.
My hindsight is not that we opened too much. It is that I would have designed openness, trust, and sustainable value together: a human-readable account, testable evidence, and machine-actionable knowledge.
Rick Anderson
First, some context for my response to this excellent question: when OA was gathering steam in the early 2000s, I was skeptical. I had (and continued to have) no problem with the idea that it’s good for people to have free access to scholarly publications; who could object to that proposition, all other things being equal? The problem, of course, is that all other things are never equal. Whenever you propose to give one person the product of someone else’s work — in this case, the work of turning research results into publishable scholarly products — at no charge, it creates a troublesome question: who, if not the end user of the product, will absorb the cost of the work that produced it?
The argument that “researchers are giving the work away to publishers for free anyway, so why is it okay for publishers to sell it back to the researchers’ institutions?” was always an unserious one, rooted in ignorance of the real-world costs of scholarly publishing. The fact that some publishers were reaping big profits on the backs of libraries was, and continues to be, a serious issue — but the OA movement was explicitly not about lowering the price of access to products of research: it was, and continues to be, about doing away with price altogether. This created unavoidable structural problems for the movement. Whenever work is done, the cost of the work will either be absorbed by the workers or passed on to someone else — and if, for some reason, that someone else must not be the person who wants to read the work (or a third-party access broker like a library), then who will it be?
One obvious possible answer was “the researcher.” This led quickly to the development of the article processing charge (APC), which is of course just another version of toll access, one that takes the toll gate from its old position between the content and the reader and plunks it down in a new position: between the researcher and the publishing service. More perniciously, the APC creates a fundamental conflict of interest for the publisher: the publisher’s interest in rejecting bad scholarship comes into conflict with its interest in preserving a revenue stream. Unfortunately, the APC is also the most obviously scalable and sustainable mechanism for directing cost away from readers, so the model took hold quickly and was enthusiastically adopted by publishers both commercial and nonprofit, some of whom were soon making more money than ever even as they were able to say (truthfully): “Look, we’re advancing open access!” This inevitably gave rise to the megajournal — and, worse, to the new phenomenon of predatory publishing, an enterprise that is basically a diploma mill for articles whereby unvetted, unreviewed content is presented to the world as peer-reviewed scholarship while the “publishers” who purvey it rake in the APC revenues. More than 20,000 of these journals are currently operating, pumping hundreds of thousands of unvetted articles (fraudulently presented as peer-reviewed research) into the scholarly ecosystem every year.
As the downsides of the APC model have become less and less deniable, there has been more and more effort invested in finding other ways of providing scholarly content at no cost to the reader. The problem is that none of them — so far — shows any serious prospect of matching the scalability and sustainability of the APC, and none has come close to achieving the APC model’s market penetration, which is currently exceeded only by that of the traditional subscription. Some of these emerging OA models (S2O, institutionally subsidized library-based publishing, crowdfunding, etc.) may yet show themselves to be durable. Others, like the preprint server, have demonstrated the ability to do some things very well both at scale and over time, but offer something very different from what is produced by formal peer-reviewed publishing.
So: looking back, what would I have done differently if I could predict the world as it is today? In the early 2000s, I was among the people talking about the downsides of APC-based OA, but what I did not anticipate was the rise of predatory publishing and the degree to which it would contaminate the scholarly ecosystem. I also failed to anticipate what a mixed impact preprint servers could have in the context of a global health crisis like the COVID-19 pandemic, during which preprints proved both a blessing and a curse. Had I been given a window on the scholarly communication ecosystem of 2026, I would have raised both of those issues with great urgency. I’m not sure doing so would have made much difference, but I’d have tried to do it anyway.
AI Use Disclosure: AI tools were permitted for drafting and editing contributions to this post. Where used, contributors remain responsible for the content, views, and final wording of their responses.
