Foundations, shell companies, trusts, banks, schools and agencies in the records
An organization is listed here because a document records something about it: money moving, a position held, a subpoena served, sworn testimony naming it. Being mentioned somewhere in the files is not enough, and every entry says which of those it is.
Receiving a donation is not evidence of wrongdoing, and most of the recipients below are charities and schools that had no way of knowing where the money came from. The evidence class on each row states what the document shows and nothing more.
A separate project maps the same release as one linked graph of people, organizations and places: epstein-data.com/entity. Its profile text is generated rather than hand-curated, as that site says itself, so it points at documents to check rather than standing as a source.
The charities he founded, funded or controlled. Six separate vehicles are now known, and what can be seen of each depends entirely on which returns survive in the files: the C.O.U.Q. Foundation's giving is itemised year by year, Gratitude America's is known for two years only, and two more have left almost no trace at all.
Corporations, LLCs and trusts used to hold his property, his aircraft and his money. The authoritative list is not a leak or a subpoena but his own estate's sworn court filing: a Verified Inventory listing every company he wholly owned, with a value for each, totalling $636,132,058 in personal property. Thirty-four of these entities were also named on a single identical exhibit attached to eight bank subpoenas.
Trust instruments signed between 1953 and 2019, with their trustees. Two wills were executed, and the trust named in each differs. Full records: the will and estate papers
Firms outside his own corporate structure that handled his travel, his accounts or his tax position, and whose records are in the files as a result.
Banks that held the accounts, moved the wires, or filed the suspicious activity reports. Full records: the banking records
Institutions with documented funding from his foundation, or with a documented member of his circle on the staff. Full records: the academic network
Schools, camps and conservatories that took his money or that his own records connect him to.
Organizations that appear as line items in his foundation's filed grant schedules.
Each of these appears on a filed schedule or a corporate record and nowhere else we have read. They keep a page so they stay findable by name; there is nothing further to show on it yet.
| Recipient | Total | Grants | Years |
|---|---|---|---|
| DARE | $10,000 | 1 | FY2004 |
| NYJTL for Cary Leeds Center | $50,000 | 1 | FY2005 |
Bodies that received foundation grants or that his correspondence places him inside.
Modelling agencies that appear in the correspondence. Presence in the files is not an allegation against an agency; the named contact and the documents are what the records show.
Grant amounts are transcribed from the page images of the C.O.U.Q. Foundation's IRS Form 990-PF filings in EFTA00224262, and each schedule adds up to its own printed total. The automatic text extraction of these scanned returns loses the column structure and pairs recipients with the wrong amounts, so no figure here is taken from it. Only schedules that have been read against the image are published; other recipients in the same filing are deliberately absent until their pages are read.
Figures transcribed from a scanned filing are checked by arithmetic before they are published: the itemised rows must add up to the total printed on the page. That is what makes them trustworthy, because the automatic text extraction of a scanned tax return misreads digits and loses columns. Where a schedule could not be reconciled, it is not published.
When a document appears in the files more than once, the copies are often redacted differently. Gratitude America's 2016 tax return is here twice: one copy blacks out the employer identification number and every figure on the page, the other shows all of them. Every figure we publish comes from the least-redacted copy we can find, and a blacked-out figure in one release is not evidence that it is unavailable.
Each entry is drawn from the dataset that already holds it, so a figure is never maintained in two places: the universities come from the academic network, the banks from the banking records, the trusts from the estate papers. The rest are recorded here.
Document counts are exact-phrase counts over the scanned text of the whole corpus, taken in a single pass. They undercount, and sometimes badly: the scanning breaks words across line ends, misreads characters, and a document can be about an organization without spelling its name the way we searched for it. A count of zero means the spelling was not found, never that the organization is absent from the files.