Summary

  • 28 CFR § 202.205 defines “bulk” with six numerical thresholds in two different units — U.S. persons for most categories, U.S. devices for precise geolocation data — measured over the preceding 12 months.
  • The rejected-offer report in § 202.1104 runs on a 14-day clock to the National Security Division, and across the § 202.105, Subpart K and Subpart L pages reviewed on 17 November 2026, no provision requires the Department to publish a licence granted, a licence denied, or an accounting of the programme.

Start with the number, because the rule does. The section headed Bulk gives six categories and six figures, and it introduces them with one sentence: “The term bulk means any amount of sensitive personal data that meets or exceeds the following thresholds” (28 CFR § 202.205). Everything after that sentence is arithmetic.

As that rendition prints the list: more than 1,000 U.S. persons for the first category, or, in the case of human genomic data, more than 100 U.S. persons; more than 1,000 U.S. persons for biometric identifiers; more than 1,000 U.S. devices for precise geolocation data; more than 10,000 U.S. persons for personal health data; more than 10,000 U.S. persons for personal financial data; more than 100,000 U.S. persons for covered personal identifiers.

One disclosure belongs beside those figures. The rendering I read elides letters inside the first category's name, printing a backtick where the interior of the word should sit. I have not quoted that line. The two figures on it — 1,000 and 100 — travel here as a reading of the section, and the linked rendition carries the text as its publisher prints it.

Two units run through one list, and the switch is the interesting part. Precise geolocation is counted in devices. The other categories are counted in people. A device is not a person, and a rule that changes unit partway down a list has made a judgement about what the harm is made of: a thousand tracked handsets is the danger the drafters saw, not a thousand named individuals deciding to be tracked.

The measurement window is a year, and it does not follow the transaction. The section aggregates across transactions involving the same U.S. person and the same foreign or covered person, with combined-data rules on top. A counterparty who buys in tranches is measured as one relationship, not as a series of small deals that each stayed under the line.

Then comes the clock. The reporting duty sits in § 202.1104, and its opening words matter more than its deadline: “A report must be filed, except as otherwise prohibited by Federal law, by any U.S. person”. The duty lands on a U.S. person that, on or after 6 October 2025, receives an offer to engage in a prohibited transaction involving data brokerage and affirmatively rejects it. The report is due within 14 days of the rejection.

What must be in the filing is where the arithmetic becomes the record. The submission has to describe the rejected transaction, the participating persons and locations, the transfer method and the supporting documentation — and the data types and volumes. The same number that decided whether the rule applied is the number the Department receives, described by the party that turned the deal down.

§ 202.1104(d) names no destination. It routes submission to Subpart L, which governs the annual report under § 202.1103 and the rejected-offer report under § 202.1104 alike. Subpart L sends both electronically to the National Security Division and requires the signature of specified officers, directors or compliance personnel, certified true, accurate and complete.

Subpart K keeps the paper for a decade. Records must remain available for examination for at least 10 years under § 202.1101, and § 202.1103 sets an annual report due 1 March covering the preceding year. The subpart's own instruction to filers is plain: “Reports required by this section must be submitted in accordance with this section and with subpart L of this part.”

And then there is the sentence that decides how much any of it is worth. § 202.105 reaches determinations, prohibitions, decisions, general or specific licences, guidance, authorizations, instructions, orders and forms, and provides that they “may be amended, modified, or revoked, in whole or in part, at any time.” A licence under this part is a permission the Department can withdraw without a stated ground, on a stated day or without one.

Now the part that is not there. Across the § 202.105, Subpart K and Subpart L pages I reviewed on 17 November 2026, I found no provision requiring the Department to publish licences granted, licences denied, or an accounting of the programme. That is a finding about those three pages as retrieved, not about 28 CFR Part 202 as a whole, and not about the Department's practice; the licensing subpart itself was not read in operative wording.

The structure around that gap can be described but not quoted. From a summarised reading of the rule that accompanied the research, and labelled here as a summary rather than as the Department's words: prohibited transactions sit in Subpart C, exemptions in Subpart E, general and specific licences in Subpart H at §§ 202.801–202.803, due diligence and audit in Subpart J, with data brokerage defined at § 202.214. None of that was read verbatim, and none of it is presented as a quotation.

Two clocks run out of this rule and they are not on the same scale. My own arithmetic: 14 days after a refusal is the rejected-offer clock; 1 March of the following year is the annual-report date. A January transaction could first appear in a filing made roughly 14 months later. The thresholds are ratios of a kind — 10,000 is ten times 1,000, and 100,000 is a hundred times 1,000 — but those are ratios between numbers printed in the instrument, not measured volumes of anything. No retained copy of the rule in the material I read carries a publication date, so none is given here.

The archive has not worked this instrument before. Two searches of BTW's published record on 17 November 2026 — one on the phrase “bulk sensitive personal data”, one on “Department of Justice”, each capped at 25 results — returned nothing, and the U.S. Department of Justice in the BTW directory counted no articles at all. What follows is the first reading here of the rule as a counting mechanism, which is a narrower subject than the geopolitics of data flows and a harder one to fake.

What would change the classification is specific enough to check against. A published licence register, with grants and denials in one list. A published count of rejected-offer reports, or of the volumes those reports described. An annual report posted rather than received. Any one of those would let a third party test the arithmetic instead of accepting it. Until they exist, the perimeter of this rule is drawn where the party it binds stops counting — and the Department's own instruments, including a licence it may revoke at any time, stay out of sight of the people whose data the count is made of.