REVIEWED · REDACTED · RELEASED
China's Acquisition & Exploitation of American Voter Data
A set of U.S. intelligence files, declassified in July 2026 and released with redactions, describes how entities tied to the People's Republic of China collected, bought, traded, and analyzed American voter and personal data across roughly a decade. This is a plain-English map of what the released text actually says — the scale, the timeline, and the stated purpose. Blacked-out specifics are left blank on purpose.
U.S. voter records in a single 45 GB file — names, ages, phone numbers, and addresses — one entry in a PRC-held catalog of "likely leaked, compromised" data.
The Paper Trail
The released documents line up into a chronology — acquisition, then possession, then analysis, then trading, then plans for what came next.
A 204.8-million-record U.S. voter file surfaces
A single 45 GB dataset of 204,822,241 voter records — names, ages, phone numbers, addresses. Alongside it: 7.42 GB of a named U.S. person's election-related emails.
Voter registration from 18 states obtained
PRC analysis of U.S. voter-registration data drawn from 18 states' midterm records, with a stated plan to run "U.S. Person Matching and Public Opinion Analysis." The North Carolina file alone held 8 million-plus voters; the Kansas file included military affiliation.
A catalog of compromised U.S. data
A PRC entity holds a list of "likely leaked, compromised" datasets: 97 entries explicitly identified as U.S.-origin, most of them bulk personal information — several described as voter-registration records. The data mostly dated 2009–2018.
U.S. intelligence takes note
A National Intelligence Council assessment finds China collecting information on U.S. candidates, campaigns, donors, and voter data, and running influence operations designed to exploit U.S. social divisions.
Notably, this assessment carried an internal dissent — an "alternative analysis" disagreeing over how far China intended to go.
Purchased, then shared — 7 states
A PRC entity had purchased 2020 U.S. voter data and shared city-level voter registration for seven states: Arkansas, Colorado, Connecticut, Florida, Ohio, Michigan, and North Carolina.
→24Next target
Eyes on 2024 — and the swing states
Discussion of targeting the 2024 U.S. elections: sharing voter-registration data, requesting "a list of swing states," and wanting to observe voting "ahead of time."
Analyst hedges in the file: "it is unknown if…," "almost certainly the 2024 U.S. Presidential election."
Declassified and made public — redacted
The files are declassified in two waves — President Trump on 3 July, Counsel to the President Warrington on 10 July — and released with names, sources, and specifics blacked out.
What Was in the Files
The 2019 catalog wasn't only voters. It read like an index of American life — sorted, volumed, and dated.
State voter rolls, by the million
- Named state file #2: 7,893,248 records — with citizenship information
- Named state file #3: 5,578,302 records (6+ GB)
- Named state file #1: 1,746,069 records — IDs, prior addresses, DOB, gender, phone
- Plus five more state databases, and a person's 7.42 GB of election emails
Enough to find a real person
- Full name, suffix, date of birth
- Home & mailing address, phone number
- Political party affiliation & historical voting records
- Polling place, military affiliation, occupation & education
Health, identity, accounts
- 28 million-record medical database — including SSNs
- A named biogenetics group; a named U.S. school
- Web & social accounts: emails, passwords, login IDs
- Business-services and web-services company records
Organizations, not just individuals
- Labor unions and a religious-media organization
- A lawyer association & a law-enforcement association
- Trade associations and think tanks
- Most entries were bulk personally identifiable information
What It Was For
The files don't leave the purpose to guesswork — they state it.
Person Matching
Tie scattered records back to specific, real individuals.
Opinion Analysis
Model U.S. public sentiment around the general election.
Target ID
"Analyze, discover the identities of important U.S. targets."
Influence Ops
Push divisive themes across TikTok, Facebook, Twitter & YouTube.
These are only the places the censors left visible. One memo references 18 states; another, 7. Most state names, and the entities that held the data, remain blacked out.
The Forensic Note
Before reading a word, we read the files themselves — the metadata. It tells its own story about how this release was built, and whether the redactions hold.
Two declassification waves days apart, with the PDFs processed 11–14 July 2026. Most files were printed and re-scanned on an office multifunction printer — each page flattened to a picture.
Yes. On the two files that kept a live text layer, the text under every black bar is deleted, not merely covered — zero recoverable names or email addresses.
One file breaks the pattern: an FBI "Albany" report generated a full year earlier (1 July 2025) by software, and titled "…Provided to Chairman Grassley."
Not the secrets — the housekeeping. Filenames like "NSA.MassagedPDB" and a preserved subject line, "NIC alternative analysis on china/election," survived the scrub.
The redaction work is sound — nothing sensitive is recoverable from under the ink. The metadata is the one part of the document nobody thought to censor.
Read This Carefully
- These are intelligence assessments, now declassified and redacted — analytic judgments, not courtroom findings.
- The 2019 "catalog" was a list of likely-leaked, compromised data aggregated from many breaches — possession of an index, not proof of a single hack by one actor.
- The 2020 assessment carried an internal dissent over China's intent; the intelligence community did not speak with one voice.
- Nothing here claims a vote was altered or an election result changed. The subject is data collection and influence — not vote tampering.
- Redactions remove the specifics — which entities, which sub-agencies, exact sourcing — so read the blanks as unknowns, not as anything in particular.