Frankfurt Appeal

Generative AI – Publishers Need Data Transparency and Effective Means of Legal Enforcement

Issued on the occasion of the Frankfurt Book Fair, this appeal calls on you, as legislators, to provide a clear statutory response to a structural problem affecting every single publishing house:

AI companies train their models on copyright-protected works without the consent of publishers and authors, exploit valuable products without remuneration, and to date have largely acted with impunity, even when unlawfully disregarding express rights reservations by rightsholders.

Despite having the technological means to do so, AI companies do not disclose which specific works are used to train generative AI models. As a result, the enforcement of reserved rights besides onerous litigation against international actors remains impossible in practice, owing to a lack of transparency. Any enforcement action is a high-risk endeavour due to the resulting evidentiary difficulties: without sufficient information, rights holders cannot establish whether their works have been used and face huge obstacles in proving potential infringements. A rights holder wishing to enforce their rights today must first undertake the burdensome task of gathering circumstancial evidence. They must then pursue often protracted, costly litigation against the world's most powerful AI corporations. For the great majority of publishers in Europe, this is not a realistic option. The result is that infringements currently carry no consequences, while revenues flow exclusively to AI companies at the expense of publishers, authors, illustrators and translators.

As legislators, you bear responsibility for putting in place a legal framework governing the development and use of AI which provides for clear obligations and liability rules. We, the undersigned publishing houses, therefore call on you to ensure the following:

• Transparency obligations for AI companies must include the disclosure of the specific training data used at work level. The guidance currently in force for AI providers falls considerably short of the requirements of the EU AI Act, whose stated purpose includes improving the enforcement of exclusive rights by rights holders.
• A licensing market with fair revenue models must be  developed. At present, AI providers are able to make use of protected works, claiming to invoke the Text and Data Mining exception, and avoid payment. Rights reservations are routinely disregarded.
• European copyright law in general must be complied with and equipped with robust means of enforcement. Rights whose infringement carries no consequences are worthless.

Signatories: