Openness and AI – Allies or Natural Enemies?

Open access was never merely a reading right for humans. Opening knowledge also opens it to new forms of searching, working and connecting. AI is therefore not a betrayal of open access – it is its stress test.

An Old Idea Meets a New Machine

The international discussion about openness and artificial intelligence has entered a peculiar phase. Two movements that depend on the free flow of knowledge are suddenly being treated as opponents. AI companies use open scholarly texts and data to build models, products and corporate value. Some advocates of open access regard this as extraction: science opens up, others profit – now through new channels as well.

These objections are not unfounded and deserve to be taken seriously. Creative Commons attribution and transparency are often missing. Automated access burdens repositories. Errors can enter models and return from there to scholarly discourse. And when a handful of companies turn open knowledge into closed systems, a new concentration of power emerges.

But it does not follow that openness and AI are natural enemies. COAR, the international association of scholarly repositories, described the two in 2026 as structurally interdependent – and explicitly recommended remaining open to benign machine use. Creative Commons likewise warns against defensive closure: more restrictive licences, CAPTCHAs and blanket bot blocks would hit not only large AI providers, but the entire ecosystem of research, archiving and translation. They would be a major setback for accessibility, which is already far from perfect.

In my view, the debate should therefore not be framed as either ‘leave everything open’ or ‘block everything’. It should start from the premise that openness in the AI world needs better rules, greater traceability and sustainable infrastructure. But openness must not deny its own purpose.

What Open Access Is For

AI has not invalidated the original arguments for open access. On the contrary, it extends their reach:

• Spread knowledge faster and accelerate research
• Enable access regardless of income, institution or origin
• Make scholarship findable, connected and reusable – by humans and machines
• Strengthen transparency, verifiability and reproducibility
• Increase the social value of publicly funded research
• Make visible languages, regions and perspectives that remain underrepresented in closed systems

The Budapest Open Access Initiative never defined open access as free reading alone. It explicitly included searching, crawling, indexing and passing texts as data to software. AI is a new and exceptionally powerful form of precisely this kind of machine use.

Open, But Not Like That?

The emotional landscape changes, of course, once open knowledge generates profit – especially for large, heavily funded companies. Yet open access has never meant a ban on commercial use. The widely used CC BY licence expressly permits it, provided attribution is given. Anyone who now argues that scholarship may be open but must not be used to train commercial models may have good political reasons. But that is a new thesis, not the old idea of open access.

This also raises an uncomfortable question. Were some leading voices always concerned above all with the best possible circulation of scholarly knowledge? Or were they sometimes more interested in fighting particular actors and their profits? If commercial success retrospectively makes a use morally suspect, science policy quickly turns into distributional politics. That debate is legitimate. But it should not pretend that openness itself opposes economic exploitation.

AI is not a betrayal of open access. It uses open content in a new environment in a way that reflects OA’s core promise: making existing knowledge easier to find, combine and use. That does not make every specific use automatically good. But machine use as a category is not inherently illegitimate.

Rules, Not Retreat

The right answer is responsible openness. Models and retrieval systems should cite sources, preserve provenance and respect licences. Providers must disclose what kinds of content they use. Repositories need protection from aggressive traffic and a fair contribution to infrastructure costs. Research needs verifiable models, documented data and people who can assess results with subject expertise.

This direction is visible internationally. UNESCO connects open science with quality, integrity, collective benefit, equity and diversity. The Royal Society calls for AI-based research to align with open-science principles. The European AI Act requires providers of general-purpose models to maintain a copyright policy and publish summaries of their training content. These are attempts to make use more responsible – not to prevent it altogether.

A Better, Richer AI

A productive alliance is already taking shape. OpenScholar makes 45 million freely accessible scholarly works available and turns them into evidence-based syntheses linked to their sources. The AlphaFold Database openly provides more than 200 million predicted protein structures, accelerating research around the world. In both cases, AI and openness reinforce one another.

Without open-access content, AI would be worse and poorer: worse because it would lack verifiable scholarly knowledge; poorer because it would rely above all on content that large companies can license exclusively. Research from smaller institutions, the Global South and less dominant languages would become even easier to overlook.

The decisive question, then, is not whether AI may use open knowledge. It is under what conditions that use serves the scholarly commons. Openness without responsibility is naive. Responsibility without openness, however, is merely a new word for closure.

Openness and AI are not natural enemies. Properly designed, they are allies – and science cannot afford to forgo that alliance.