The problem of copyright protection for authors when training artificial intelligence (AI) models is becoming increasingly deep. A large number of lawsuits against companies developing AI tools prove this. This issue requires an immediate response from both legislative bodies and publishers and development companies. However, the approaches to granting permits for training AI models differ.

Recently, it became known that The Copyright Clearance Center (CCC) has decided on a new approach to collective licensing of content use in internal AI systems. The license allows users to effectively access an agreed set of rights from copyright holders and provides for the payment of royalties for the use of their works. At the same time, the academic publisher Wiley announced agreements with technology firms to provide access to the content of its authors for training large language models without their prior consent, noting that “Wiley authors are remunerated for licensing their work based on their ‘contractual terms’.

Of course, the journal business operates on a completely different model. Increasingly, authors are paying for the distribution of content by paying for the processing of open access articles. In the area of remuneration for their publishing work, it is understood that scientists usually see the benefits of authorship through the secondary outcomes of publication. These include increased reputation and recognition, but most importantly, professional advancement.

Recognition of authorship as a prerequisite is enshrined in the Budapest Declaration, which states that “the only role of copyright in this area should be to give authors control over the integrity of their work and the right to proper recognition and citation.” The theory is formulated as follows: since authors usually do not receive remuneration for their contributions, their work should be distributed under a CC BY license to comply with the concept of “open access”. Even if AI absorbs their work for learning purposes, the requirement to attribute the author remains for this content.

A key problem with increasing the secondary market for academic text licenses by expanding licenses to include these texts in AI tools is the potential to disrupt the reputation ecosystem that supports scholarly publications. Much of the secondary market for researchers' remuneration depends on citations. However, if the attribution market is a key factor in motivating authorship, and AI tools create content and provide search results without attribution, this could become a serious problem. Publishers could use the licenses they negotiate to push AI tool developers to promote the inclusion of citations in their tools.

Perhaps the best strategy would be to include a requirement for citation and attribution in content licenses. This is an important opportunity for the publishing community to engage AI developers and insist that attribution be part of any future technical development of AI tools.

We can only hope that publishers will take the interests of authors into account when negotiating and approving licenses with technology companies.