
For astir a year, Tyler Johnston has been collecting nationalist accusation astir the interior workings of OpenAI, and for the past month, he’s been moving connected a study to assistance the nationalist recognize and visualize it.
That report, dubbed The OpenAI Files, is retired today. It’s a collaboration betwixt the Midas Project and the Tech Oversight Project, 2 nonprofit tech watchdog organizations, and it’s billed arsenic the “most broad postulation to day of documented concerns with governance practices, enactment integrity, and organizational civilization astatine OpenAI.”
The interactive site, which amounts to much than 50 pages and implicit 10,000 words, chronicles OpenAI’s improvement from nonprofit probe laboratory to moneymaking household name, and the information concerns and imaginable conflicts of involvement it’s generated on the way. It uses sources similar firm disclosures, ineligible complaints, unfastened letters, media reports, and much — things that The Verge and different quality outlets person often covered arsenic idiosyncratic stories, but gathered successful 1 spot here. Charts and information visualizations assistance exemplify what we cognize astir OpenAI’s interior workings, including the nonprofits’ champion conjecture astatine OpenAI’s firm structure, its archetypal capped-profit structure, and the projected program for its restructuring.
One salient taxable is however overmuch OpenAI executives and committee members basal to straight oregon indirectly summation from the company’s successes. The nonprofits see their champion conjecture astatine CEO Sam Altman’s ain concern portfolio, which includes a laundry database of companies — like Retro Biosciences, Helion Energy, Reddit, Stripe, Rewind AI, and Rain AI — that presently person overlapping concern with OpenAI, specified arsenic partnerships, vendor relationships, oregon imaginable acquisition talks.
Johnston, who is enforcement manager of The Midas Project, told The Verge that the task is partially astir pointing retired “the ways successful which the imaginativeness that they had successful the precocious 2010s departed from the mode that they’re behaving contiguous successful 2025.”
The nonprofits down the study said they received nary funding, assistance, editorial direction, oregon “support of immoderate kind” from Elon Musk, xAI, Anthropic, Meta, Google, Microsoft “or immoderate different OpenAI competitor” and that the study has “complete editorial independence.” OpenAI declined to comment.
“We’re successful an archival task here, wherever we’re showing what OpenAI was similar then, what they’re similar now… We’re conscionable putting that accusation successful beforehand of the scholar and asking the scholar to gully their ain conclusions astir what to marque of it,” Johnston told The Verge.
You tin work the study astatine OpenAIFiles.org.