Open access varies significantly between natural sciences and the humanities and social sciences. Photo: TUCHONG

On Feb. 16, the Royal Society confirmed on its official website that, with libraries supporting its Subscribe to Open (S2O) model, eight of its world-class subscription journals will be published as open access (OA) in 2026. These include the world’s oldest peer-reviewed journals, Philosophical Transactions A and Philosophical Transactions B, and authors will be exempt from article processing charges (APCs). This means that any author or reader can publish and read papers free of charge, no longer limited to users of subscribing libraries. This development has once again brought OA to the forefront of academic attention.

Different types of OA models

As a widely recognized, efficient, and equitable model for transforming publishing, S2O is centered on the annual transition of journals from traditional subscription-based models to OA. In simple terms, OA publications refer specifically to peer-reviewed research papers that readers can access online free of charge, without subscribing to journals or paying fees, removing barriers to the dissemination of research findings.

Depending on the level and timing of access, OA publications can be classified into several types. Gold OA refers to journals that are fully open access by policy, with all papers freely available. Hybrid OA and Bronze OA are other journal-based models: Hybrid OA journals publish a mix of closed and open access articles, while Bronze OA journals make articles conditionally available, such as after an embargo period.

Another model, Green OA, denotes articles published in closed access journals but self-archived by authors in independently open repositories for public access. It should be noted that some journals require authors to pay APCs for Gold, Hybrid, or Bronze OA publications, but this is not universal and has not become an industry-wide standard.

Uneven development

In August 2025, the National Center for Science and Engineering Statistics (NCSES), a subsidiary of the US National Science Foundation, analyzed global patterns of OA publication output and scientific impact from 2003 to 2022 using Scopus, a bibliometric database of publication metadata, which includes the OA status of research publications. The findings show that since 2003, the number and share of OA articles have risen to the point that OA has shifted from an infrequent type of peer-reviewed publication to accounting for nearly half of global publications in 2022.

In an interview with CSST, Peter Baldwin, a distinguished research professor of history at the University of California, Los Angeles, said that progress has been steady and largely in the right direction, though it varies significantly between the hard sciences and the humanities and social sciences.

According to Baldwin, the sciences increasingly adopt gold or diamond OA models, incorporating costs into research grants. However, he warned that reductions in federal funding, limits on overhead for federal grants, and taxation of university endowments in the United States could constrain overall research funding and, in turn, weaken support for diamond models. “That is true for all research spending, not just on making content OA,” he added.

At the same time, Baldwin noted that these developments do little to address the issue of excessive profits among scientific publishers. “In the humanities and social sciences, there is nothing like this,” he said. Journals and periodicals in these fields, he explained, remain largely subscription-based, and OA applies only sporadically.

He further observed that attempts by libraries to negotiate agreements with publishers—allowing faculty to publish OA for a flat or discounted fee—do little to resolve the broader challenges, whether in expanding access or addressing the cost structure and profit margins of academic publishing.

Baldwin also pointed out that the humanities and social sciences face an additional challenge: books rather than articles. “Costs are higher, but more intermittent of course, and who is going to pay?” he asked.

He noted that some presses have sought to form library buying consortia to support OA publishing. For example, MIT Press has brought together a group of publishers to share the costs of publishing selected new books, as well as parts of their backlists, openly accessible. “Again, this is good, but it doesn’t solve the problem across the board,” Baldwin said.

OA in age of AI

Regarding the impact of AI, Baldwin said that many of the challenges it raises are not specific to publishing, libraries, or access to works, but affect society more broadly and will need to be addressed at that level.

“For example, the tendency for Large Language Models (LLMs) to fabricate and make up things, the vast increase in AI-generated content of highly variable quality, and the likelihood of declining content quality as LLMs feed on and ‘learn’ from content that they themselves have generated—the so-called ‘enshittification’ process,” he noted.

Despite these challenges, he also noted that AI may offer potential solutions to some of these issues in OA. LLMs can generate summaries of works, potentially at varying levels of detail. A summary that is 99% accurate might technically avoid constituting a copy and thus fall outside copyright restrictions. Readers may find such summaries accessible, legal, and useful, although this has yet to be tested in court.

Baldwin also argued that AI can help identify the portions of works most relevant to a reader’s interests. This could make readers more willing to rely on LLM-generated summaries, especially when paired with more targeted access to specific sections of works within the limits of fair use. Much will depend on how broadly fair use exemptions can be applied to deliver content that remains under copyright. He noted that this may advantage US-based institutions able to rely on a comparatively expansive interpretation of fair use, as well as libraries over commercial entities.

LLMs have already absorbed most legally available online content and are now seeking access to high-quality materials held in major research libraries. Yet even if they can use such materials for training, they cannot necessarily provide direct access to readers. Baldwin noted that related court cases are still pending.

Truly global OA will take time

Baldwin said that although there have been positive developments in OA, with more journals and publishers paying attention to publishing costs, progress remains uneven and incremental. “Copyright law and the economic model of publishing still jointly stand in the way of comprehensive solutions,” he noted.

In his view, for works still in print and under copyright, gold or diamond OA models remain necessary, though both limit participation by authors without adequate funding. For works that are out of print but still under copyright—which make up a large share of holdings in major research libraries—some solutions are gradually emerging.

“US libraries are trying to use the exemptions provided by Section 108 of US copyright law for library lending to scan, preserve, and provide access to these out-of-print but still in-copyright works,” Baldwin said.

At the global level, some countries have adopted digital legal deposit systems requiring works to be submitted to national libraries, ensuring that once they enter the public domain, they can be made openly accessible. Baldwin noted that, by contrast, “in the United States, recent judicial rulings have weakened the Library of Congress’ ability to require deposit of works, whether digital or print—a clear step backward.”

“It will be a long time before truly global OA is achieved,” Baldwin said, noting that if reforms such as expanded fair use interpretations or digital exemptions take hold and are upheld in court, they could, in principle, enable broader access to such works.

Even so, cost considerations will remain a constraint. “If there is to be global access, there will probably also have to be some form of global distribution of costs,” Baldwin added.

He further pointed to Wikipedia as an example. Even such a successful open knowledge project exhibits clear geographic imbalances—the English-language version is far larger than those in other languages such as Arabic.

“Wikipedia has attempted to develop a universal machine language to translate articles from different language versions into a common semantic form, thereby expanding the coverage of smaller languages. But to some extent, this is still just spreading content from larger languages, leading to homogenization of global knowledge rather than genuinely expanding diversity,” Baldwin said.

He argued that a similar logic applies to OA: “If the world can eventually read all the books in the New York Public Library, that will be good, but it will just be spreading the knowledge already in that library, not adding, expanding, or improving that knowledge. Truly global OA requires a reciprocal movement of information across the globe, not just a one-way flow.”