As of 2026-08 a US firearm sale may proceed three business days after the background check begins even if the check never finishes — of 556,496 NICS denials across 2008 through 2015, the ATF referred 558 subjects for prosecution and US Attorneys Offices accepted 254, fewer than 32 a year

What is happening?

Under federal law a licensed firearms dealer may complete a transfer once three business days have passed since a background check was initiated, if the National Instant Criminal Background Check System has not returned a denial in that time. The transfer is lawful whether or not the check ever finished. The provision is 18 U.S.C. § 922(t)(1)(B).

A check that resolves as a denial only after the firearm has already changed hands is called a delayed denial. From that point the correction runs backwards: the FBI contacts the dealer to establish whether the transfer actually took place, and then refers the matter to the Bureau of Alcohol, Tobacco, Firearms and Explosives for retrieval. In 2025 the FBI referred at least 2,232 purchasers to that agency for firearm retrieval.

The window has been changed once, and only for part of the buying population. The Bipartisan Safer Communities Act of 2022 gave the FBI up to ten business days for buyers aged 18 through 20 when juvenile or mental health records show a potential disqualifier. For buyers 21 and older the three day default is unchanged, and for buyers aged 18 through 20 the sale still proceeds once the longer window expires.

The unresolved share is large. Of the checks that ran past the three day mark in 2024, about 71 percent were still without a final determination from the FBI.

Whose problem is this?

The institution that sets the length of the window is not the institution that has to retrieve a firearm after the window closes, and no source opened here describes a mechanism that ties the two together. That separation is the shape of this problem rather than a detail of it.

Where does this problem end?

The boundary here is procedural rather than substantive, because the question is not who may lawfully buy a firearm but what happens while the system has not yet answered that question. Every value dispute named in the right column sits next to this problem without being inside it.

What is the state now, and what should it be?

Now

Needs a new measurementthe target state: no source opened here names a federal target for how many delayed checks should be resolved, how fast they should be resolved, or how many firearms transferred under the default should be recovered. The only numbers carrying binding force are the two statutory windows themselves, three business days in general and up to ten business days for the narrow class added in 2022. Bills introduced in 2019, 2021 and 2025 would lengthen the window further, but a bill that has not become law states a preference rather than a standard, and no opened source supplies an agency target to put in its place.

How big is it?

Between about 3,490 and 3,963 firearms a year, on two averages that do not belong on one line. One is 3,490 a year across 1998 through 2014, taken from a reported total of 55,887 firearms for that span. The other is 3,963 a year across 2014 through 2018. They come from different producers using different methods over windows that meet at one edge, so the pair marks two bounds rather than two points on a series.

A third figure counts people instead of firearms and covers a single recent year. In 2025 the FBI referred at least 2,232 purchasers to the ATF for retrieval after a delayed denial. A referral is not a confirmed transfer and not a completed retrieval, so this number cannot be set on the same axis as the two averages, and no opened source performs that reconciliation.

Neither the averages nor the referral count includes the checks that were never resolved either way, and in 2024 that group was about 71 percent of the checks running past three business days. There is no way from these sources to say how many of those unresolved checks involved a prohibited purchaser.

Volume and exposure move together. In March 2020 monthly checks reached 3.7 million, an increase of 1.1 million over the same month a year earlier, and that month produced at least 35,000 potential default proceeds of which at least 523 reached prohibited purchasers. A surge does not change the length of the window, it changes how many checks fail to fit inside it.

None of these three figures is a count of people over a comparable window, and no source opened here supplies one. The affected population for this document is therefore recorded as not derivable rather than estimated, and the numbers above stand as bounds and single year observations rather than as a total.

Under what conditions does it arise?

1. The default runs toward completion of the sale. The statute does not hold the transfer until an answer arrives. It sets a period after which the absence of an answer is treated as permission, which means the party carrying the risk of an unfinished check is not the party running it.

2. Correction has to run backwards. Once the firearm has changed hands, the remedy is no longer a refusal at the counter but a retrieval by a federal agency, followed in some cases by a referral for prosecution. Each of those steps requires locating a person, and each can fail without anything being recorded as a failure of the check.

3. Volume compresses the window without changing it. The three day period is fixed in days, not in capacity, so any period of elevated purchasing pushes a larger absolute number of checks past the mark. March 2020 is the clearest measured instance.

4. The prosecution tail is narrow. Across 2008 through 2015 the ATF referred 509 denial cases covering 558 subjects and US Attorneys Offices accepted 254 of those subjects, fewer than 32 a year, in a period when more than half a million NICS transactions ended in a denial.

5. The one extension applies to a narrow class and still ends in a proceed. The 2022 change reaches buyers aged 18 through 20 with a specific kind of potential disqualifier, and when its longer clock runs out the transfer may go ahead exactly as before.

What has been tried?

Every legislative attempt has taken the same shape, which is to make the window longer rather than to require that the check be finished before the transfer. That is a change in how much time the system gets, not a change in what happens when the time runs out.

What was found?

Why is it still unsolved?

Institutional gap — the statute already answers the question of what to do when a check is unfinished, and the answer it gives is that the sale may go ahead.

The default is a choice about who carries the delay, and it cuts both ways. The same clause that allows a transfer without a determination is the clause that keeps an unresolved check from becoming an open ended hold on a lawful purchase. Moving the line moves the cost from one set of purchasers to another, which makes it a legislative decision rather than an administrative one, and legislative decisions in this area have reached a floor vote twice in seven years without reaching law.

Correction after the fact is more expensive than refusal before it, and the record shows the size of that difference. Between 2008 and 2015 more than half a million transactions ended in a denial, 558 subjects were referred for prosecution and 254 were accepted, fewer than 32 a year. Retrieval is a separate track from prosecution and its own completion rate is not published in any source opened here, so the backstop can be described in referrals but not in outcomes.

The pathway also produces very little record of itself. A transfer under the default looks the same on paper as a transfer after a clean check, because the dealer has followed the law in both cases. When a delayed check ends in no determination at all, which was the outcome for about 71 percent of the over three day cases in 2024, there is no denial to record and nothing to retrieve. A process that generates no event generates no count, and a gap with no count is hard to argue about in either direction.

That last point cuts against overstatement as much as against complacency. Without published retrieval completions, a claim that the backstop works and a claim that it does not are both unfalsifiable from the public record as it stands in 2026.

What observation would mean it is solved?

Candidates — (a) the statute is amended so that a transfer may not proceed until the check returns a determination (b) the share of checks that run past the window without a final determination falls and stays down (c) the FBI and the ATF publish annually how many delayed denials were referred and how many firearms were actually recovered, and the distance between the two narrows.

(a) alone is weaker than it looks. Lengthening the window moves the deadline rather than removing it. The 2022 change did exactly that for one age band, and when the longer clock expires the sale still proceeds. Any amendment that ends in a proceed by default is the same structure at a different hour.

(b) alone can move for reasons unrelated to the problem. The unresolved share is a ratio whose denominator swings with purchasing volume, and March 2020 shows how far it can swing inside a single month. A quiet year lowers it and a surge raises it without anything changing in how checks are resolved.

(c) alone counts the backstop rather than the exposure. A perfect retrieval record would still mean the firearm was transferred first and recovered afterwards. But it is the measurement that does not exist today, and without it the first two observations cannot be read against each other at all.

What is it connected to?

Fills with researchthe completeness of the state records that feed the system, private transfers that never reach a check, state waiting periods running alongside the federal window, and the prosecution policy that decides which denial cases are charged. Relation type and evidence grade were not established in this round.

What these sources do not say

• The annual volume of checks. No source opened here gives the total number of NICS transactions for 2024 or 2025, so the shares above cannot be converted into counts for a recent year.
• How many firearms are actually recovered. No source opened here reports retrieval completions. The 2025 figure counts referrals, and the 2008 through 2015 record covers prosecution referrals and acceptances, which measure a different step.
• Whether the estimates belong on one line. The 1998 through 2014 average, the 2014 through 2018 average and the 2025 referral count come from different producers using different methods and different units. None of them acknowledges the others and none states a shared counting basis.
• What became of the 2025 bill after introduction. No source opened here records a committee action, a floor vote or a Senate companion for it.
• What the 2025 staffing changes did to retrieval. The report carrying those figures covers agent assignments and inspection capacity, and does not report retrieval counts or the time taken to recover a firearm. The connection is not drawn by the source and is not drawn here.
• Any published agency position on the unresolved share. No source opened here carries a statement from the FBI or the ATF about the 71 percent figure or about what becomes of those cases. The 2016 Office of the Inspector General report is the nearest thing to an agency account of the denial process, and it predates these figures by roughly a decade.

See the evidence

No primary federal document was read in full. The statutory text was read on a commercial republication of the US Code rather than on the official code, and the Office of the Inspector General figures come from the release page for that report rather than from the report itself. The 2024 operational report from the FBI, the two congressional record pages and the sponsor office page all returned HTTP 403 to automated retrieval, which is why the annual transaction volume, the vote record for the 2019 bill and the status of the 2025 bill are absent above rather than estimated. Where the opened sources overlap they agree: the three day default appears in the statutory text and in every secondary source, and the structure of a delayed denial as a referral for retrieval appears in both the 2016 inspector general material and the 2025 policy page. Where they do not overlap the gap is left visible. The two multi year averages of firearms reaching prohibited purchasers, 3,490 for 1998 through 2014 and 3,963 for 2014 through 2018, come from a news analysis and an advocacy research report respectively, neither cites the other, and neither states a counting basis that would let the two be joined. Both are graded medium for that reason, and neither is carried in the title without its span. This is a Path A output, a research based definition, so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range Not derivable

The reason and what is missing are listed under “What is missing” below