For seven weeks I’ve been building this model act on the bones of one statute: Title III, the 1968 wiretap law. I’ve borrowed its architecture section by section — its definitions, its prohibition-first default, its authorization tiers, its use restrictions. So a fair reader might expect that when I finally turn to the government’s other great surveillance statute — the Foreign Intelligence Surveillance Act, and especially its Section 702 — I’d mine it for parts too. I won’t. This statute intentionally does not emulate FISA, and particularly not FISA 702. FISA 702 isn’t the model. It’s the warning.
Here’s the distinction. Title III and 702 are both surveillance statutes, but they are built on opposite instincts about the moment that matters. Title III put its safeguards at the front — at capture — and made the later use of what it captured require a fresh trip to a judge. Section 702 does close to the reverse: it collects in bulk, retains in bulk, and then lets the government search the retained store — and the fight over 702 is, at bottom, a fight over that search. Set aside the debate over the foreign-targeted collection itself; that’s not where the American Fourth Amendment problem lives. The problem lives at the query.
And to see why the query is the problem, you don’t need a critic’s characterization. You can use the FBI’s own. The Bureau’s public FISA Query Guidance — a redacted, unclassified-in-part document from DOJ’s National Security Division — is admirably plain about what a query is. The querying standard, it says, governs queries of “raw FISA collection,” and it defines that phrase in a footnote: “content and noncontent information, including metadata.” It then explains that the government’s minimization procedures govern “acquisition, retention, and dissemination” of FISA-acquired U.S.-person information — and “the querying of raw FISA collection.” Read those two together and the machine comes into focus. When an analyst runs a query, the query runs against the raw store — the unminimized collection, everything that came in. Minimization is not a filter on what can be searched. It’s a set of rules about what may be kept and shared after the fact.
That is the exact inversion of the statute I’m building on, and it’s why 702 is the anti-model. Title III minimizes at the front: the agent monitoring a wiretap had to stop capturing a call the moment it turned personal and non-pertinent, so the store was filtered on the way in. Section 702 does no such filtering at the point of search. The query reaches the whole raw corpus — including the incidentally-collected communications of Americans that the minimization rules are supposed to protect. The safeguard moved from “what may be searched” to “what may be kept,” which is the wrong end of the pipe, because by the time the keep-or-discard rules operate, the search has already happened. A protection that applies only after the search is not a protection against the search.
Now put that next to the case law, because a court has already seen it. In United States v. Hasbajrami, the Second Circuit held in 2019 that querying the 702 database is a separate Fourth Amendment event from the original collection — reaching into the retained store is its own search, judged on its own terms, not automatically blessed by the lawfulness of the collection that filled it. On remand in 2025, the district court took the next step and held that a U.S.-person query requires a warrant. That is the whole thesis of this blog, stated by a court, in the highest-stakes surveillance context we have — and now you can see exactly what it’s a search of: the raw, unminimized store, in DOJ’s own words. The query is the search. The query is a new use of retained data. And it is a search of everything.
This is not hypothetical, and the scale is the tell. In a stretch between 2020 and early 2022, the FBI ran more than 278,000 queries of Section 702 data that did not comply with the rules governing U.S.-person searches. Not 278 — 278,000. That is the predictable failure mode of a query-first regime whose only real filter sits downstream: when the search reaches raw collection and the discipline is a standard rather than a structural bar, the searches run into the hundreds of thousands, and a great many of them run outside the lines. Over-collect, retain, re-search — the pattern this blog has traced through Ganias, Richman, and Nejad — here at the scale of a standing national database.
I want to be precise about what the law already requires, because it cuts in my favor. The 2024 reauthorization did add one narrow checkpoint: under 50 U.S.C. § 1881a(f)(2), before the FBI may access Section 702 content returned by a U.S.-person query that was run solely to find evidence of a crime, in connection with a predicated criminal investigation not related to national security, it must first obtain an order from the FISA Court — the only exception being a reasonable belief that the content could help avert a threat to life or serious bodily harm. So a court-order requirement already operates, today, in the narrowest and most criminal-justice-like slice of 702 querying. Hold onto that, because it detonates the loudest objection to reform: that requiring a court’s sign-off before searching Americans’ data is unworkable and would blind the intelligence community. It plainly isn’t unworkable — the government does it now, in the lane closest to ordinary law enforcement. The reform fight is simply about extending that same principle to the rest of the U.S.-person queries, and the coalition pushing it is genuinely cross-ideological; the last broad attempt failed in 2024 by a single vote, 212–212.
Let me still give the national-security case its due, because it isn’t nothing. Speed matters in counterterrorism; a foreign target’s American contact may be the thread that unravels a plot, and a warrant requirement before an analyst can check a known threat’s domestic connections could, in a true emergency, cost time. The collection was lawful. And the government argues the 2024 reforms have already reduced the abuses. These are serious points, and a serious statute answers them — with a bounded exigency exception, not with no rule at all. That’s the same answer § 7 gave on Tuesday to “you can’t unsee it,” and § 6 gave to “deletion is dangerous.” You protect the ticking-bomb case without blessing the routine, suspicionless searching of a raw store that produced a six-figure count of violations. And the deepest objection — “the collection was lawful, so the query is fine” — is exactly the inference Hasbajrami rejected and this whole project exists to reject. Lawful possession is not lawful search. Lawful collection is not a standing license to query.
So when I say the model act does not emulate FISA 702, here is what I mean concretely. The model act would not let the government hold a raw, unminimized store and search it at a standard’s discretion; it minimizes and segregates at the front, the way Title III did and § 4 and § 5 require. It would not treat the reasonableness of collection as authorization for later search; § 7 makes a new use of retained data require fresh judicial authorization — the second-warrant rule — as a structural bar, not a downstream reporting nicety. And it would not bury the one real check, a court order before searching a citizen’s data, in the narrowest possible lane; it would make it the rule. Title III is the blueprint because it put the guard where the intrusion is. Section 702 is the counter-model because it moved the guard downstream and let the search run free. Every section of this statute is, in part, an argument against becoming 702.
Next Tuesday, § 8: disposition — the end of the lifecycle.
Leave a comment