Ask whether the buyer keeps, resells or publishes what it takes: record buyers range from open-source archives to resellers
OpenAI's data partnership offers two routes, an "Open-Source Archive" that "would be public for anyone to use in AI model training" and "Private Datasets" for its own models, and its page names no payment for either. Mode says it de-identifies the agreed copy "before reselling it for AI research and training", and Polyshares says "We anonymize it, build it into data sets, and supply AI labs". At the other end Labelbox's Prism makes exclusive the default (held clause labelbox--exclusivity) and Grepped says data is "never resold to competitors or published" (held clause grepped--resale). Corpus Bridge's worked example is a 3-year, non-exclusive licence.
Dated 22 Oct 2026Source checked 22 Oct 2026Quote found on its page again 26 Oct 2026Basis: indicative
“Open-Source Archive: We’re seeking partners to help us create an open-source dataset for training language models. This dataset would be public for anyone to use in AI model training.”
OpenAI: OpenAI Data Partnerships (Internet Archive capture of 28 Dec 2023), checked 9 Nov 2023
Mode (BizBuyData): home page, checked 22 Oct 2026
Everything read from this source (1)Archived copy, 16 Oct 2026
Polyshares: home page, FAQ, checked 22 Oct 2026
Everything read from this source (10)Archived copy, 16 Oct 2026
Corpus Bridge: home page, illustrative example, checked 22 Oct 2026
Everything read from this source (15)Archived copy, 18 Oct 2026
- Lead with what is not on the open web: that is what the labs say they pay for
- Scraped profile data carries a deletion risk that company records do not: never take it into a sale
- A company's operating record (Slack, email, tickets, CRM, files), bought by Polyshares
- Polyshares and company owners: price range, company operational records (Slack, email, tickets, CRM, files)
- Polyshares: what it buys and pays
- Polyshares: When you are paid
It is shown on its own, not counted into a figure.