204 Million Voter Records: What the Declassified China Files Actually Say

A declassified tranche describes how PRC-linked entities collected, bought, and analyzed American voter data for a decade. We read the text — and the metadata. Here's what's really in the files.

Start with one number.

In a single 45-gigabyte file, dated 2016, sat 204,822,241 American voter records — names, ages, phone numbers, addresses. It was one line item in a catalog of “likely leaked, compromised” data that an entity of the People’s Republic of China was holding onto.

That file is one of two dozen documents the White House declassified and released, in redacted form, in July 2026. The set carries a blunt government title: “China’s Acquisition and Exploitation of American Voter Data.” We went through all of it — and then we went through the files themselves, the metadata, because with a release like this that turns out to matter too.

Here’s the honest version, in plain English.

The inventory: what’s actually in the files

The centerpiece is a 2019 catalog. A PRC entity was sitting on a list of leaked and compromised datasets. Most entries were targets in other countries — but 97 were flagged as explicitly American, the bulk of them raw personal information. Several of those datasets were described, in so many words, as U.S. voter-registration records.

The voter data isn’t vague. The catalog itemizes eight named state voter databases. One held 7,893,248 records — with citizenship information attached. Another, roughly 5.5 million. A third listed voter IDs, previous addresses, birth dates, gender, and phone numbers. Sitting beside them: a 28-million-record medical database that included Social Security numbers.

A separate memo describes PRC analysis of voter registration pulled from 18 states, drawn from midterm-election records. The North Carolina file alone covered more than 8 million voters. The Kansas file went further — it carried military affiliation, historical voting records, and voters’ occupation and education, which the document itself calls “a priority.”

The fields they kept read like a wish list: full name and suffix, date of birth, home and mailing address, phone number, party affiliation, where you vote, and how you’ve voted before.

The chronology: the timeline is the story

Lay the documents end to end and you get an arc, not a snapshot:

• By 2016, that 204-million-record voter file already exists.

• In 2018, PRC analysts are working 18 states’ worth of midterm voter data, with a stated plan to run “U.S. Person Matching and Public Opinion Analysis.”

• By 2019, it’s all sitting in a catalog of compromised American data.

• In 2020, U.S. intelligence takes formal note: China is collecting information on American candidates, campaigns, donors, and voter data.

• By 2023, a PRC entity has purchased 2020 voter data outright and is sharing city-level voter records for seven states — Arkansas, Colorado, Connecticut, Florida, Ohio, Michigan, and North Carolina.

• Heading into 2024, the discussion turns to the next election: sharing more voter data, requesting “a list of swing states,” and wanting to observe the voting “ahead of time.”

Collected, catalogued, analyzed, bought, sold — and pointed at the next cycle.

The purpose: what it was for, and what it wasn’t

The files don’t leave you guessing. In their own language, the goals were person-matching (tie the records back to real people), public-opinion analysis on U.S. elections, and identifying “the identities of important U.S. targets.” A 2020 assessment adds the influence layer — pushing divisive themes on race, COVID, immigration, and guns across TikTok, Facebook, Twitter, and YouTube.

Now say what this isn’t. This is mass data collection and influence — not vote-flipping. Nothing in these files claims a single ballot was changed or a result altered. Anyone who tells you the documents prove otherwise hasn’t read them. The real finding is unsettling enough on its own: a foreign government built a detailed, queryable picture of the American electorate, person by person — and kept adding to it.

The Briefing: the whole release on one page

Scale, timeline, the data taxonomy, the named states, and our metadata audit — the full breakdown on a single page. Open the interactive version →

The forensics: we read the documents, not just the headlines

This is the part Rational Ground exists to do. Before trusting a word of the text, we looked at the files as files.

A few things jump out. The release came in two waves days apart — declassified by President Trump on July 3, and by Counsel to the President on July 10 — with the PDFs assembled over a handful of days in mid-July. Most of them were printed and re-scanned on an ordinary office multifunction printer, which flattens every page to a picture. That’s the fingerprint of a manual, last-mile declassification job.

We also checked whether the redactions actually hold — because that’s usually where a release like this springs a leak. They hold. On the two files that kept a live text layer, the text under every black bar is deleted, not merely covered; nothing sensitive is recoverable. Whoever did the redacting did it right.

One document breaks the pattern: an FBI “Albany” report generated a full year earlier and titled, right in its file properties, “…Provided to Chairman Grassley.” And the housekeeping leaked what the censors didn’t — filenames like “NSA.MassagedPDB” and a preserved email subject line, “NIC alternative analysis on china/election,” survived the scrub. The metadata is the one part of a document nobody thinks to censor.

The fine print: read it honestly

Two cautions, because they matter. These are intelligence assessments — analytic judgments, now declassified and redacted — not courtroom findings. And the intelligence community did not speak with one voice: the 2020 assessment carried an internal dissent over how far China actually intended to go. The redactions also strip the specifics — which entities, which agencies, exact sourcing — so read the blanks as unknowns, not as blanks to fill in with your priors.

None of that softens the core of it. A foreign adversary spent years acquiring the voter file of the United States — by the hundred million — and, heading into 2024, was still shopping for more.

---

About the source. Drawn from “China’s Acquisition and Exploitation of American Voter Data,” a set of 24 U.S. documents declassified and released in redacted form, July 2026. Every figure here is quoted from the released text; blacked-out specifics are left blank. Representative files: 200M Voter Records Compromised · 18 States Memo · PRC US Voter Data — 7 States (2023) · PRC Target 2024 Election (2023) · NIC — Foreign Threats to 2020 US Elections · Summary (Parts 1–3).