Mustafa Sulaiman, head of Microsoft’s AI division, has made a big difference with his recent report titled Using Web Content to Train Language Models on Authorship. It sparked a heated debate. Accordingly, he opined that anyone can freely use publicly accessible content on the Internet. This view raises legal and ethical questions.

AI: Is internet content free to use?

Speaking to host Andrew Ross Sorkin, Suleman supported the practice of using public internet content to train AI models. Asked whether big companies have stolen the world’s intellectual property rights, he replied: “I think that content on the open Internet has been subject to fair use since the 1990s. Anyone can copy, reproduce and reproduce it. It’s freeware if you want to.” This statement reflects a broad interpretation of the “fair use” principle, which is interpreted differently in the United States and Europe.

EU Future Networks Director: “Data security should not be an innovation killer”

Suleiman spoke of a “grey area” and called for a clearly formulated “opt-out” option. Websites, publishers or news organizations may openly oppose the use of AI. In ongoing litigation, Microsoft and OpenAI have been accused of using copyrighted material to train AI models. With this in mind, Suleiman’s position is particularly explosive.

Microsoft AI CEO Mustafa Sulaiman: A social contract for content on the open web is “freeware” for training AI models. pic.twitter.com/FN1xrqnJC0

— Tsarathustra (@tsarnick) June 26, 2024

Legal uncertainties and gray areas

In the United States, copyright protection arises automatically when someone creates a work. Publication on the Internet does not constitute a waiver of these rights. The rules are similar in Europe: by publishing a literary, scientific or artistic work, the author automatically receives copyright protection, which applies from the moment of creation and does not require a formal registration procedure. This also applies to works published on the Internet.

The concept of “fair use” allows the use of copyrighted works under certain conditions, but how this regulation can be applied to the use of such content to train AI models remains controversial.

Sulaiman’s reports clearly show that big tech companies like Microsoft do not take legal regulations seriously and do not use publicly available data from the Internet. Questions about the legal situation in this regard will keep courts around the world busy for a long time.

New Google data protection guidelines: AI will be allowed to learn our content in the future