AI data licensing deals, July 2023 to October 2026: a public dataset and what it shows

Abstract

Licences of content to developers of artificial intelligence (AI) are negotiated privately and recorded in no registry; what is known comes from announcements, securities filings and press reports. This paper describes an open dataset that compiles those disclosures, with a source link, date and verbatim quote for every row, and reports what it shows. The version analysed here (reviewed 17 October 2026) holds 111 records dated 11 July 2023 to 15 October 2026. It names 24 licensees; OpenAI appears in 29 records, Meta in 14 and Google in 12. News publishers are the licensor in 52 records (46.8%) and music companies in 18. The use licensed shifted: 66.7% of 2024 records include training, against 36.1% in 2025 and 44.7% in 2026 to date, while licences to show content in AI answers became the most common news arrangement. Disclosure of money is sparse and skewed. Nineteen records (17.1%) carry a dollar figure, but only 11 of the 89 records that describe a specific licence do, and 7 of those 11 rest on press reports citing unnamed people. No music licence in the table has a stated amount. A contract term is stated in 7 records. We describe the fields, collection and dating rules, versioning and verification, and the disclosure and survivorship biases that limit inference. The dataset is released under CC BY 4.0, with a snapshot on Zenodo.

Keywords: AI training data, content licensing, data licensing, generative AI, disclosure bias, news publishers, music licensing, open dataset