AI, Open Access and Shared Responsibilities: Reflections from Open-Access-Tage 2026 

Anna Wałek

Wed 21 Oct 2026

---

When we first started discussing the plan of organising a workshop on AI and Open Access, one thing became clear very quickly: no single organisation can see the whole picture. 

At OAPEN and DOAB, we have been grappling with questions around AI crawlers, content reuse, discoverability, metadata enrichment, and the growing pressure that automated traffic puts on scholarly infrastructure. At the same time, our colleague Franziska Deller from Barbara Budrich Verlag was encountering a different side of the same transformation through editorial workflows, publishing policies, authorship, and quality assurance. What struck us was that these conversations were often happening separately. 

Publishers were discussing AI-generated manuscripts, editorial responsibility, and transparency. Repositories and infrastructures were struggling with bot traffic, platform sustainability, and machine access to scholarly content. Libraries were exploring the implications of AI for discovery and research support, while researchers themselves were trying to understand what responsible AI use means in scholarly practice. 

Yet all of these questions are deeply interconnected. 

That observation became the starting point for our workshop, “AI in Open Access Publishing: Governance, Responsibility and Practical Decision Making,” at Open-Access-Tage 2026 in Linz. 

Together, we wanted to bring these perspectives into the same room. To broaden the discussion, we invited Kathrin Ganz (FU Berlin/Open Gender Journal), bringing the perspective of a journal editor and researcher, and Julia Boltze-Fütterer  (Kooperatives Bibliotheksnetzwerk Berlin-Brandenburg), who shared experiences from the repository and infrastructure landscape through her work with OPUS. Together with the OAPEN perspective represented by Anna Wałek and Hanna Varachkina, this created a conversation spanning publishing, repositories, infrastructure, libraries, and research communities. 

What made the workshop particularly interesting was that participants quickly moved beyond discussing AI as a technology. 

Instead, the discussion focused on something much more practical: What decisions can organisations no longer postpone? 

We need AI policies. The question is what they should contain. 

One of the strongest messages emerging from the discussion was that many organisations have moved beyond the stage of simply observing developments in AI. They now need policies. But participants quickly discovered that saying “we need an AI policy” is much easier than deciding what that policy should address. 

Several themes repeatedly surfaced: 

The discussion revealed that organisations are not only facing technical questions. They are increasingly being asked to make governance decisions. 

Who is responsible when AI is involved in a workflow? What should be disclosed? And what level of transparency is both realistic and meaningful? 

Academic integrity means more than plagiarism now 

A particularly interesting discussion centred on scholarly integrity. For years, conversations about research ethics and publication quality have focused heavily on plagiarism and citation practices. AI is forcing organisations to broaden that discussion. Many participants argued that the idea of independent scholarly work needs to evolve. Authors should increasingly be expected to disclose the use of AI tools when preparing articles, books, or other scholarly outputs. Transparency was seen as more valuable than attempting to detect AI use after the fact. This was especially important given a widely shared concern: current AI detection tools are often unreliable and difficult to use as a basis for editorial decisions. 

Several participants also pointed out a practical reality that is often overlooked. Every new requirement creates additional work. Checking AI disclosures, verifying references, and assessing AI-generated content requires time and resources. If organisations introduce new requirements, they also need to recognise the impact on editorial workloads and budgets. 

Kathrin shared a concrete example. Faced with increasing numbers of AI-generated and hallucinated references, reference checking has moved from a final quality-control step to an earlier stage of the editorial process, allowing problematic submissions to be identified before significant resources are invested. 

Policies are not enough if contracts do not support them 

An important part of the discussion focused on the relationship between publishers and authors. 

One point resonated strongly across the room: Publishers cannot reasonably expect authors to follow requirements that are not reflected in publishing agreements and policies. 

If organisations expect authors to disclose AI use or comply with specific standards, these expectations need to be clearly communicated and formally supported. At the same time, participants recognised that contracts alone will never solve the problem. Good scholarly practice remains essential. Policies and agreements can establish expectations, but trust, integrity and professional responsibility remain at the heart of scholarly communication. 

Can reviewers upload manuscripts to AI systems? 

Few topics generated as much immediate agreement as this one. 

Participants broadly agreed that unpublished articles, manuscripts and book proposals should not be uploaded into public AI systems. Authors, reviewers and editors often have confidentiality obligations. Unpublished scholarly work contains intellectual contributions that have not yet entered the public domain and may involve sensitive information. While some participants acknowledged the potential role of secure and controlled AI environments, many highlighted the lack of clear guidance and the need for organisations to establish practical rules for handling unpublished content. 

Open Access infrastructures face a different AI challenge 

The second half of the workshop shifted attention from AI-generated content to something less visible but equally important: AI-generated traffic. Julia Boltze-Fütterer shared experiences from the OPUS repository environment, which immediately resonated with challenges reported by other infrastructures and repositories. Interestingly, several participants were unfamiliar with the scale of automated traffic generated by AI-related crawling and data collection. 

This led to a broader discussion about questions that repositories increasingly have to answer: 

• Which bots should be allowed? 

• Which should be restricted? 

• How much automated access can infrastructures realistically sustain? 

• How can organisations remain open while protecting services and preserving meaningful usage statistics? 

Participants exchanged experiences with solutions including Anubis, Cloudflare and other approaches to traffic regulation and infrastructure protection. What became clear is that bot management is no longer simply a technical issue. It is a governance issue. Every repository and infrastructure now faces decisions about what openness means in an environment where both humans and machines consume scholarly content. 

Are existing licences prepared for the age of AI? 

No discussion about AI and Open Access can avoid the question of content reuse. For decades, Open Access has been built on enabling access and reuse. AI challenges us to think more carefully about what those principles mean when content can be collected, analysed and incorporated into AI systems at unprecedented scale. Participants discussed whether existing licensing frameworks provide sufficient clarity and protection. 

Creative Commons licences remain fundamental to Open Access, but they were not designed specifically for AI-driven reuse. Questions emerged about whether organisations can realistically control such reuse through licences alone and whether entirely new legal frameworks may eventually be needed. 

Emerging approaches such as opt-out mechanisms and Creative Commons Signals attracted considerable interest as potential building blocks for future solutions. At the same time, participants repeatedly returned to a difficult reality: Even the most carefully designed legal or technical framework will only be effective if it is respected by those using the content. 

From debate to decision-making 

Perhaps the most important outcome of the workshop was not any single conclusion, but a shift in perspective. The conversation suggested that the Open Access community has largely moved beyond asking whether AI will affect scholarly communication. That question has already been answered. The challenge now is deciding how organisations should respond. Publishers, repositories, libraries, infrastructures and researchers may encounter AI in different ways, but they increasingly face the same underlying questions about transparency, responsibility, sustainability, governance and trust. 

The workshop did not attempt to provide definitive answers. Instead, it created a space where different parts of the scholarly communication ecosystem could compare experiences, challenge assumptions and identify common concerns. Participants also expressed interest in continuing the discussion beyond the workshop through dedicated AI working groups. 

What emerged was not a blueprint, but something equally valuable: a recognition that these challenges are shared and that meaningful solutions will require collaboration across the Open Access community. 

And perhaps that is the most encouraging conclusion of all.