Last week’s post ended on an uncomfortable fact: the modern search happens at analysis, on a complete copy the government is holding at its leisure. This week is about that copy — where it came from, and the question nobody’s law answers: how long does the government get to keep it?
Two acts are bundled together in the word “retention,” and the model act pulls them apart. The first is copying. When the government executes a warrant on a device, it doesn’t extract the responsive files; it images the whole drive, producing a bit-for-bit clone — a new artifact that didn’t exist until the government created it. I made the case in the very first post that this copy deserves to be its own defined object, and that the Constitution takes copying seriously enough to name it twice — “papers” in the Fourth Amendment, the exclusive right to control the copying of one’s “Writings” in the Copyright Clause. The copy is not a neutral convenience. It is the thing that makes everything downstream possible: the indefinite hold, the re-search years later, the mining of the non-responsive remainder. No copy, no retention problem.
The second act is keeping it, and here the law simply runs out. There is no retention clock. Once the government has lawfully imaged a drive, essentially nothing requires it to ever delete the copy — not when the search is done, not when the case ends, not when the defendant is acquitted, not in ten years. The default, in practice, is keep it forever. And “in practice” is doing real work in that sentence, because the keep-forever default isn’t written anywhere as a rule; it’s what you get when no rule says otherwise and every institutional incentive points toward hoarding.
I’ve spent five weeks documenting what that default produces. It’s the retained mirror images in Ganias, held for years and then mined for a new crime. It’s the copies in Richman that a court finally had to order the government to give back. It’s the entire body of email in Nejad, loaded into a database and searched six hundred times. And it’s the plain reality of federal practice: whatever the records schedules say on paper, what goes into the official case file is kept, and kept, and kept. The retention problem isn’t a series of scandals. It’s the ordinary condition of digital evidence, and the scandals are just the times it happened to surface.
So § 6 supplies the clock and the guardrails the current non-system lacks. Four pieces.
First, a retention clock for the non-responsive remainder. This is the heart of it. Once the government has completed its authorized search and segregated the responsive material, the non-responsive remainder — the overwhelming majority of any forensic image, the intimate 95 percent the warrant never reached — must be deleted on a defined deadline, unless a court affirmatively orders it preserved. Retention stops being the default and becomes a decision someone has to make and justify. The clock runs; when it expires, the remainder goes, unless a judge has said otherwise for a stated reason.
Second, segregation, carried over from § 4 and § 5. You cannot run a retention clock on “non-responsive data” unless the government has actually separated the responsive from the non-responsive — the failure that sank the government in both Ganias and Richman. So § 6 depends on the segregation duty the earlier sections build: identify what the warrant authorized keeping, and the rest is on the clock.
Third, a handling record. Every copy the government makes — every duplicate, every transcript, every derivative — and every time it opens or searches the image, gets logged: a chain of custody for data, in whatever form the content takes. Forensic examiners already keep chain-of-custody records for physical evidence; no one thinks that’s an unreasonable burden. Section 6 extends the same discipline to the copy and its analysis, for the same reason: so that what was retained, and what was done with it, is provable rather than a matter of the government’s later say-so. Remember that the scope violation in Nejad stayed hidden for years precisely because no record surfaced it. A handling record is what makes a retention clock enforceable instead of aspirational — you can’t audit a deletion deadline you have no record against.
Fourth, a sealed reference copy with a neutral custodian, for the cases where legitimate future need is real. Because it is sometimes real: appeals happen, post-conviction claims surface, and the innocence interest that drives our biological-evidence preservation laws is genuine. The answer is not to keep everything, forever, in the investigating agency’s own hands. It’s to preserve a single sealed copy — hashed, integrity-locked — with a neutral custodian such as a court, reachable again only on a fresh showing to a judge. This is exactly the remedy the Richman court had to invent on the spot: delete the working copies, escrow one sealed copy with the court rather than the Justice Department, require a warrant to touch it again. Section 6 makes that the default architecture instead of a judicial improvisation a defendant has to win.
The nearest thing Title III has to any of this is the sealing rule, § 2518(8)(a) — and it says more about retention than I first suggested. It requires wiretap recordings to be sealed under judicial direction, protected from alteration, kept for at least ten years, and destroyed only upon a court’s order; the presence of that seal, or a satisfactory explanation for its absence, is a prerequisite to using the evidence at all. So Title III did not ignore retention — it built a regime around it. But notice what kind of regime: a preservation mandate. The wiretap era’s fear was the government quietly destroying or doctoring the recording, so the statute sealed it, kept it for a decade, and put destruction in a judge’s hands. That is nearly the opposite of a deletion clock.
That isn’t a contradiction of § 6; it’s the key to it. Title III could safely mandate preserving the recording because the recording was supposed to be minimized — an agent had to stop capturing non-pertinent calls in real time — so the sealed tape was, in theory, all responsive, and preserving it protected the evidence and the defendant’s ability to challenge it, not a warehouse of innocent life. The forensic image has no such front-end filter; it is the whole life, unminimized and overwhelmingly non-responsive. Port § 2518(8)(a) straight across and you would mandate keeping every imaged drive, under seal, for ten years — a preservation rule turned into a hoarding rule, precisely because minimization no longer runs at the front. So § 6 splits an object Title III never had to split. The responsive evidence and the sealed reference copy get the full Title III treatment — sealed, integrity-locked, destroyed only on a court’s order, preserved as long as appeal, authentication, and the defendant’s own access genuinely require. The non-responsive remainder gets the inverse, because it is exactly what minimization used to exclude and cannot anymore: an affirmative deletion clock. Section 6 is not a betrayal of the wiretap model. It is that model’s purpose — preserve the pertinent, protect its integrity, and never let the government destroy evidence on its own say-so — carried into a world where the pertinent no longer arrives already separated from everything else.
There’s a public window into how that preservation runs, and it corrects an instinct worth naming. The FBI’s electronic-surveillance recordings sit on a NARA-approved disposition schedule that, for Title III collections, directs destruction “10 years after last date of interception, after securing any necessary authorization for destruction from a court of competent jurisdiction, or upon closure of the corresponding investigative case file, whichever is later.” It is tempting to read that as a retention limit that never really limits — the clock runs from the last interception, so a fresh interception years later resets it, and “whichever is later” defers even that to a case closure the Bureau in practice never reaches. But that reading misses who the preservation is for. The person recorded benefits from it: those recordings are, by design, his own pertinent communications, minimized at capture, and keeping them available means he can use them — to challenge the government’s account, to supply the exculpatory context around a clipped quote, to authenticate what was actually said. Preservation of the pertinent recording is a feature, and it is the very feature § 2518(8)(a) exists to protect. The long retention of the recording is not the problem.
The problem is what the recording rule leaves untouched — and this is the gap in Title III I care about most. Section 2518(8)(a) governs the recording; the NARA schedule governs the media — the discs and the cassettes. Neither governs the copies made from those recordings, and neither governs the transcriptions. Title III seals the original, keeps it for a decade, and requires a judge’s order to destroy it — and then says nothing about the duplicate a prosecutor makes, or the transcript that goes into the case file. So the identical content — the words spoken on the call — comes to live in two places under two opposite regimes: 1. the sealed, court-controlled original; and 2. a freely-copyable transcript in a file the Bureau keeps indefinitely, subject to none of the sealing, custody, or judicial control that guards the tape.
That gap is the whole thesis of this project, surfacing in the retention context. What the Fourth Amendment protects is the content — what was said, what was written — not the medium that carries it, and not the tone of a person’s voice. A statute that locks down the audio recording while letting a verbatim transcript of the same words circulate untracked has guarded the container and released the contents. Section 6 closes that by attaching its rules to the content in every form it takes — the original, every duplicate, every transcript, every representation. Retention, destruction, and the handling record follow the content, not the cassette. If the protected thing is the content, then a copy of the content is the content, and a transcript of the content is the content; a regime that guards only the first recording of it is guarding the least of what matters.
Now the objection, and it’s the most emotionally powerful one in this whole project. Deleting evidence is dangerous. Appeals and habeas petitions can run for years; exoneration claims surface decades later; related investigations develop; and the government has independent legal duties to preserve records. Force deletion, the argument goes, and you will destroy material that later turns out to matter — sometimes material that would free an innocent person — and you will hand the guilty a way to run out the clock. Better, on this view, to keep everything, because you never know what you’ll need. And courts have blessed exactly this instinct: in United States v. Aboshady, the First Circuit saw nothing unreasonable in the government holding all of the data it had pulled from a set of email accounts until the defendant’s criminal appeals were complete.
I take the innocence point seriously — it’s the reason § 6 has a sealed-reference-copy provision rather than a flat delete rule. But look hard at “you never know what you’ll need,” because it is precisely the general-warrant instinct the Fourth Amendment exists to defeat. “Keep it all in case it’s useful later” is the justification for every dragnet ever built; it proves too much, because it justifies retaining everything about everyone forever. The exoneration interest doesn’t require warehousing the entire digital life of every person whose device was ever imaged; it requires preserving the responsive evidence and a sealed, auditable copy against the day a court says it’s needed. And the “deletion is dangerous” framing has the risk backwards. An ever-growing, indefinitely-held store of citizens’ most intimate data is not safe. It is a standing liability — the honeypot that gets breached, the database an insider abuses, the copies that get re-searched for purposes no one authorized. Retention is not the cautious choice. It’s the risky one; we’ve just been counting only one side of the ledger.
Aboshady rewards a close read, because it names some real interests before it reveals the gap. Leaning on Ganias‘s own reasoning, the First Circuit acknowledged that keeping a complete mirror of seized data can be genuinely necessary — to preserve the evidence’s integrity, to authenticate it at trial, even to let the defendant himself challenge it — and that separating responsive from non-responsive data is not always feasible. Those are real reasons to preserve, and I won’t wave them away. But watch how the court traveled from “preservation can be necessary” to “keep all of it until the appeals are done.” It did not hold that indefinite retention is constitutionally sound. It held that because the warrant set no time limit on retention, it could not call retention-until-appeals unreasonable — “given the absence of any such time limit,” as it put it, it would not supply one. That is not a defense of the keep-everything default; it is a diagnosis of it. With no rule to enforce, the court deferred, permitting the government to hold the entire database — all 430,000-odd documents — until “the appeals are completed,” an endpoint it otherwise left undefined. Section 6 is the rule that warrant lacked, and it is built to honor the very interests Aboshady named without the pathology it tolerated: integrity, authentication, and defense access are exactly what a sealed reference copy preserves. What they do not require is leaving the whole working dataset in the investigating agency’s own hands, indefinitely, on a clock that reads “whenever we’re done.” Preserve the responsive evidence and a sealed copy for as long as justice genuinely needs them; put the non-responsive remainder on a defined schedule. Aboshady is a court with no tool to draw that line. I argue to hand it one.
Copy, and the clock starts. That’s the whole of § 6: the government may make the copy it needs to do the search but holding that copy is a decision with a deadline, a record, and a narrow, court-supervised exception. Not a permanent entitlement that attaches the moment the imaging finishes. Next Tuesday, § 7: use and disclosure — what the government may do with what it found, and the second-warrant problem that runs underneath it. And Thursday, the same retention-and-requery fight this post is about, playing out right now at the largest scale we have: the battle over Section 702.
(Model statutory text for § 6 comes in Pass 2 this fall. This post is the why; the drafting is next.)
Leave a comment