The Standoff That Keeps Everything · Two Laws, Opposite Orders 

Two bodies of federal law give the government flatly opposite instructions about the data it takes from you. One is a law about letting go. The other is a law about holding on. When they meet over the same forensic image, and neither clearly wins, the government does the thing that violates neither: it keeps everything, indefinitely. That outcome — permanent retention as the accidental byproduct of a legal standoff — is what this post is about, because it’s the records-law face of the retention problem this blog keeps circling.

Start with the law about letting go. The Federal Records Act, on the books since 1950, governs how federal agencies create, keep, and get rid of their records. Its animating premise is disposal on a schedule: an agency may destroy a federal record only under a disposition schedule the National Archives has approved, and those approved instructions are mandatory — NARA even publishes disposal notices in the Federal Register for public comment before shorter retention periods take effect. Read the FRA and you’ll find it is fundamentally a statute about systematic disposal — routing inactive records to destruction on a timetable and permanent ones to the Archives. By its own design, records law wants the government to let go.

Now the law about holding on. Criminal evidence has to be preserved — for trial, for appeal, for post-conviction review, and, for some categories, essentially forever. Federal law requires the preservation of biological evidence in federal cases for as long as the defendant is incarcerated, precisely because innocence claims can surface decades later. Constitutional duties to preserve exculpatory material run alongside. Evidence law, left to itself, wants the government to hold on.

For most of the history of both regimes, they rarely collided head-on, because physical evidence is discrete: a gun, a ledger, a bag of powder, each individually preserved or scheduled. The forensic image breaks that. When the government images a phone, it creates a single new object that is, at once, evidence (the responsive files the warrant sought) and an enormous mass of non-evidence (the intimate remainder, plus the data of every third party the target ever spoke to). Records law looks at that object and says: schedule it and dispose of it. Evidence law looks at the same object and says: preserve it. And neither regime was built for the thing in front of it — records schedules were written around paper case files, and evidence-preservation rules don’t distinguish the 5 percent the warrant covered from the 95 percent it didn’t. So the image lands in a gap, and the institutionally safe move in that gap — the one that breaks neither the ban on unauthorized destruction nor the duty to preserve — is to keep all of it. On paper that reads like two commands canceling out. In practice it is worse than a tie, because one of the two barely fires at all — and the agency that holds most of this data is the proof.

If that sounds abstract, it was concrete two weeks ago. When the judge in Richman ordered the government to give back the copies of the professor’s data, the government’s independent obligations to preserve records were part of the tangle — you cannot simply order an agency to destroy material it may be separately required to keep. The court’s fix — delete the working copies, escrow a single sealed copy with a neutral court, require a warrant to touch it again — was in part an improvised reconciliation of exactly these two regimes: preserve what must be preserved, relinquish the rest, and put a neutral party between the government and the data in the meantime. A judge hand-built, for one professor, the rule that ought to exist for everyone.

The FBI is the proof, because it holds most of this data and it does not, as a practical matter, let records go. Whatever the schedules say on paper, the Bureau’s operative default is permanence: the investigative record — the FD-302 that memorializes an interview, the EC that documents a step in a case, the case file that collects them — is retained indefinitely, for periods so long they amount to the same thing. There are real institutional reasons for that, and a long history behind it; an agency’s memory is one of its assets. But it means the “dispose” half of records law never truly operates here. And the seized forensic image isn’t sorted into some separate bin marked for deletion — it’s swept into the same case file, held by the same institution, under the same reflex to keep. An agency that will not delete its own paperwork is certainly not going to delete a full image of a bystander’s phone from a case that closed years ago. So it sits, and the sitting, multiplied across every device the Bureau images, becomes a vast and growing digital holding that no schedule was ever going to empty. That is the records-law version of the copy that never gets deleted — and the reason it never gets deleted is that, here, almost nothing does.

It’s worth noticing that records law is under unusual public scrutiny right now — NARA has been reminding agency leaders of their retention duties, and there’s been sustained congressional attention through 2025 and into 2026 to federal records being disposed of, altered, or lost. But nearly all of that attention runs the opposite direction from this post: it’s about officials destroying records they were supposed to keep — vanishing messaging apps, contested disposals. The mirror-image failure — the government keeping citizens’ seized data it arguably should dispose of — draws almost none of the same energy, even though it’s the identical statutory machinery jammed in reverse. We worry loudly about the government deleting its own records and hardly at all about it hoarding ours.

There’s even a tell buried in the FRA itself. Since the 2014 amendments, the law has directed NARA to write standards for digitizing records “with a view to disposing of the source records” — the government has a whole framework for the situation where the copy becomes the record and the original is destroyed. The law knows how to reason about copies and originals when it wants to. It has simply never written the rule for the seized forensic image: make the copy, and then — nothing. The silence is the gap.

Let me put the strongest case for the current muddle. The two regimes, someone will say, aren’t really in conflict if you read them carefully: evidence-preservation duties are bounded by the life of the case and its appeals, records schedules can be written to match, and the “keep everything forever” result is a mere implementation failure, not a command of law. And erring toward preservation has a real virtue — the biological-evidence statutes exist because wrongful-conviction claims surface decades later, so a bias toward keeping protects the innocent, not just the prosecution.

The exoneration point is the serious one, and the model act should honor it — but it argues for preserving the responsive evidence and a verifiable record of what was done, not for warehousing a complete image of everyone’s entire device on the theory that someday some byte might matter. You can fully protect the innocence interest with a sealed, escrowed reference copy and a handling record while still deleting the non-responsive mass on a clock. And “it’s just an implementation failure” gives the game away: a legal standoff that reliably produces the wrong result in practice is the paradigm case for legislation. You do not fix it by trusting agencies to voluntarily schedule and destroy material they have every institutional incentive to keep. You fix it by writing the schedule into law.

Which is what § 8 of the model act has to do. Disposition can’t be left as the residue of two other regimes — one pushing to preserve, the other too weak in practice to make anything go away. It has to be an affirmative rule: define the forensic image and the non-responsive remainder as objects with their own disposition schedule; require deletion of the remainder on a clock unless a court affirmatively preserves it; keep the responsive evidence as long as justice genuinely requires; and supply, by default, the neutral-custodian escrow that Richman had to invent from scratch. Tuesday’s post governed how the government takes your data. This is what happens to it afterward — it falls into the space between a law that is supposed to make the government let go and a law that makes it hold on, and because the letting-go never actually happens, holding on wins by default every time. The model act’s job is to force the choice the current system never makes.