European Parliament's position on the Artificial Intelligence Act would require public summaries of copyrighted training data

2023-06-14EU Policy

European Parliament, Amendments adopted on 14 June 2023 on the proposal for a regulation laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), P9_TA(2023)0236, on COM(2021)0206, adopted by 499 votes to 28 with 93 abstentions

The European Parliament adopted its negotiating position on the proposed Artificial Intelligence Act by 499 votes to 28 with 93 abstentions, and the Commission, the Council and the Parliament now begin trilogue negotiations on a final text. The Parliament's text adds a category of "foundation model," a model trained on broad data at scale for a wide range of downstream tasks, and attaches obligations to its providers. For foundation models used in systems intended to generate content, Article 28b(4)(c) of the Parliament's text requires the provider to "document and make publicly available a sufficiently detailed summary of the use of training data protected under copyright law." The text expands the prohibited practices to real-time remote biometric identification in publicly accessible spaces, predictive policing, emotion recognition in law enforcement, border management, the workplace and education, and untargeted scraping of facial images from the internet or closed-circuit television footage to build facial recognition databases. It adds to the high-risk list systems intended to influence voters in political campaigns and the recommender systems of very large online platforms. None of this is law; the three institutions must agree a single text before the regulation is enacted, and the copyright summary obligation is one of the provisions the Parliament added that the Commission's 2021 proposal did not contain.

What it changesA duty to publish a "sufficiently detailed summary" of copyrighted training data would put into the public record information that today serves two purposes in a patent program: it is held as a trade secret, and it is written into the specification as the support for claims that recite training on a particular corpus. If the obligation survives trilogue, a company offering a generative model in the Union will be disclosing, in a public summary, data choices it may also be relying on as a trade secret, and the secret ends when the summary is posted. The decision to make now is which training details the company is willing to describe in a European filing and which it would rather keep out of every document, because a patent application that describes the corpus in detail and a trade secret program that depends on the same corpus cannot both be right. The Parliament's text is an opening position and the clause may change before the regulation is final.

More on AI regulation and policy · Copyright and training data

← All recent developments · Search all developments

Burbage Law is the practice of Jon Burbage, a registered patent attorney. 342 United States patents have issued from applications he wrote; the full list is on the Patents page. More about Jon Burbage.

This is general information about a published decision, agency guidance or piece of legislation. It is not legal advice and does not create an attorney-client relationship. Items concerning offices outside the United States are reported for information only and are not counsel on the law of those countries.