The forfeiture fund at 28 U.S.C. 524(c) is available to the Attorney General without fiscal year limitation — money that arrives that way never passes through the annual appropriations bargain

What is happening?

In 32 states and the federal system, law enforcement agencies keep between 90 and 100 percent of civil forfeiture proceeds, in funds those agencies themselves control. Another 12 states award between 45 and 80 percent. Six states and the District of Columbia — Maine, Maryland, Missouri, New Mexico, North Carolina and Wisconsin — direct all of it to neutral funds instead. That is the map as of 2026-03-24, the publication date of the fourth national census of the program.

The flow is large and it is measured only in dollars. At least USD 82 billion in property has been forfeited nationwide since 2000 — USD 57.4 billion federal and about USD 25 billion state and local — on data running through 2023. Federal forfeiture revenue in 2023 was USD 4,495,470,000, of which two matters together accounted for USD 1.87 billion. State-law forfeiture in 2023 came to just under USD 300 million across 45 states and the District of Columbia. Net assets of the federal forfeiture fund rose from USD 766 million in 2000 to over USD 5 billion in 2023.

Most of it never reaches a judge. Across 2000 through 2023, 71 percent of the forfeitures of the Department of Justice that produced revenue were completed administratively — decided inside the seizing agency, with no judicial review. The median currency forfeiture under state law is USD 1,678 across 24 states for 2019 through 2023, against an estimated USD 3,300 to hire a lawyer for a straightforward case.

Whose problem is this?

The body that receives the money is the body that chooses the target, and no step between the two requires a neutral decision-maker. That arrangement is not hidden and it is not disputed. Both the concurrence and the dissent in the 2024 Supreme Court decision described it, and the Court still declined to require a preliminary hearing.

Where does this problem end?

The boundary matters because two reforms that sound alike do different things. A state can abolish civil forfeiture and keep the money, and a state can give the money away and still have its agencies drawing on the incentive through Washington. Montana is the first case and North Carolina is the second. Only one of the two axes changes what an agency stands to gain.

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

Now

Needs a new measurementthe target state: no source opened here names a number, a threshold or a date against which this program could be scored. The recommendations published with the census — remove the financial incentive for civil and criminal forfeiture alike, provide prompt post-seizure hearings, raise the standard of proof, close the equitable sharing loophole, require full transparency — name no figure and no deadline. No statute sets one and no agency publishes one. The fourth edition also does not compare its own findings with the 2010, 2015 or 2020 editions, so the national census of this program does not state whether the thing got better or worse.

How big is it?

The unit here is a property-taking event, not a person. No source in this file publishes a count of people who lost property, and none publishes a count of properties taken — the entire national account is denominated in dollars. The count that does exist sits on the other side of the same transaction: 707,735 equitable sharing requests were decided between 2000 and 2023, of which 652,903 were granted. The requests made by the agencies are counted one by one. The owners are counted in currency.

Floor — about 27,000 events a year. 652,903 granted requests across 24 years is roughly 27,200 a year. This counts only the equitable sharing slice, about one-eighth of the USD 82 billion national total, and excludes every state-law forfeiture and every federal forfeiture not shared with a state or local agency.

Ceiling — about 390,000 events a year. Just under USD 300 million of state revenue in 2023, divided by the state median of USD 1,678, gives about 179,000. Federal deposits of USD 14.44 billion across 2019 through 2023, about USD 2.89 billion a year, divided by the DOJ median currency forfeiture of USD 13,695, give about 211,000. Dividing a total by a median overstates a count whenever the distribution has a long right tail, so both figures are mathematical ceilings rather than estimates.

The width of that band is not sampling noise. It is the unknown shape of the value distribution, and nobody publishes that shape. Two matters were 42 percent of federal forfeiture revenue in 2023, which says the federal skew is severe and the true federal count sits nearer the low end. Any narrowing requires an extrapolation assumption, and such an assumption belongs stated out loud rather than folded into a midpoint.

Events are not people and the error runs in both directions. One person can lose several properties, several people can claim one property, several agencies can file requests against a single seizure, and one request can cover several assets. No source publishes the mapping. Property seized and later returned appears in no count at all, and neither do the households that depended on a single seized vehicle.

Under what conditions does it arise?

1. The appropriation is permanent and does not lapse. The fund established at 28 U.S.C. 524(c) is available to the Attorney General without fiscal year limitation. Money that arrives that way never passes through the annual appropriations bargain, which is the ordinary mechanism by which a legislature prices one enforcement choice against another. Net assets of that fund rose from USD 766 million in 2000 to over USD 5 billion in 2023 with no legislature voting on the growth, and 44 of 51 jurisdictions run a state analogue of the same arrangement.

2. The forum is chosen by the party that benefits. Administrative forfeiture, decided inside the agency, was 71 percent of the DOJ forfeitures that produced revenue across 2000 through 2023. The Institute for Justice reports research on DOJ data for 1998 through 2019 finding property returned in fewer than 8 percent of administrative forfeitures against nearly 37 percent of civil-judicial ones. Same agency, same statute, same property types — the variable is whether a neutral decision-maker looked.

3. The taking is priced below the cost of contesting it. The median state currency forfeiture is USD 1,678 and a lawyer for a straightforward case is estimated at USD 3,300. Claim windows run 20 to 30 days, some states require a bond, and in half the states the statutory deadlines to reach a hearing sum to 188 days. Claim rates run from 6 percent in New Jersey to 34 percent in Indiana, representation reaches 6 percent in Arizona and 7 percent in Oregon, and defaults exceed 70 percent in Kansas and South Dakota. Non-contest is not consent but arithmetic.

4. A state prohibition becomes a routing decision. Equitable sharing returns up to 80 percent of proceeds to the participating state or local agency, and the realized share averaged 65 percent from 2000 to 2023. DOJ granted 92 percent of all requests across that span and 99 percent of them when funds were available, approving the exact share asked for in 78 percent of requests. When a state lowers the take of its own agencies, it raises the relative return of the federal route rather than lowering the return.

5. The federal channel is controlled by a memorandum. A 2015 order banned DOJ adoption of property seized by state and local agencies, and adoptions fell from 26 percent of equitable sharing forfeitures in 2014 to 5 percent in 2016. A 2017 order reinstated adoptions with conditions, Treasury followed, and adoptions stabilized near 10 percent from 2019 onward. Nothing further has moved in the nine years since.

What has been tried?

Two of these worked and the rest did not touch the mechanism. The change at the IRS removed one category of seizure and Congress made it durable. New Mexico closed both the state stake and the federal channel, and reported no forfeitures under state law at all from 2019 through 2023. Everything else reformed the procedure, bounded the size, reallocated the flow or graded the states. The fund statute is unamended.

What was found?

Why is it still unsolved?

Incentive inversion — the body that decides what to take is funded by the decision, and every reform in the record has changed the terms of the taking rather than the funding of the decider.

Start from what the ordinary control on enforcement looks like. An agency that wants to do more of something asks a legislature for the money, and the legislature prices that request against everything else it could buy. Forfeiture removes the request. The fund at 28 U.S.C. 524(c) is available without fiscal year limitation, so the money neither lapses nor passes through the bargain, and the balance of that fund grew from USD 766 million to over USD 5 billion between 2000 and 2023 without a single vote on the growth. This is the object of the problem and it is the one thing no reform in the record has touched. The 2000 statute reformed the procedure. Timbs bounded the size in 2019. Two attorneys general reallocated the flow in 2015 and 2017. Culley declined to add a hearing in 2024. The fund statute is unamended.

The second movement is that the cheapest path for the agency is the one nobody adjudicates. The median state currency forfeiture is worth about half of what a lawyer costs, the claim window runs 20 to 30 days, some states require a bond, and in half the states the statutory path to a hearing runs 188 days. Owners respond the way the arithmetic dictates — claim rates of 6 to 34 percent, representation of 6 to 7 percent in the two states measured, defaults above 70 percent in two more. It would be possible to read that silence as consent if the outcome data did not exist, and it does exist. Property came back in fewer than 8 percent of administrative cases and nearly 37 percent of the ones a court decided. The forum that the seizing party may choose is the one in which property comes back least often.

The third movement is what makes a state reform individually reversible without anyone reversing it. Equitable sharing returns up to 80 percent to the participating agency; DOJ granted 92 percent of requests over 24 years, 99 percent of them when funds were available, and approved the exact share asked for in 78 percent. So when a state lowers the take of its own agencies it does not lower the return, it changes the route. North Carolina is the demonstration. It has only criminal forfeiture in most cases, it sends every dollar of state forfeiture proceeds to public schools, and its agencies still drew USD 364,458,695 from Washington between 2000 and 2023, with an estimated 57 percent of the law enforcement agencies in the state certified for the program. New Mexico is the control. It closed the same door in July 2015 and reported no state forfeitures at all across five straight years. The difference between the two is one clause about equitable sharing money — not the abolition, and not the grade. On the federal side that clause is held by a memorandum: one order closed adoptions in 2015, one reopened them in 2017, and the pen belongs to the office that receives the money.

What observation would mean it is solved?

Candidates — (a) the number of jurisdictions in which law enforcement receives neither state forfeiture proceeds nor equitable sharing proceeds rises above 2 of 51 (b) the share of forfeitures decided by a neutral adjudicator rises and the administrative share falls (c) the rate at which owners contest and recover property rises.

(a) alone counts statutes. North Carolina shows what a statute can leave standing: it removed the state stake completely and its agencies drew USD 364 million through the federal program anyway, and six threshold states barred sharing below a dollar line while 4,415 below-threshold assets were shared regardless. A jurisdiction count moves when a legislature votes. The money moves when a route closes. Those are not the same event.

(b) alone can be produced without anyone deciding differently. Administrative shares move with the mix of cases — criminal forfeiture reached 51 percent of DOJ forfeitures in 2023 on the strength of two large matters, which was composition and not reform. A falling administrative share would also say nothing about whether the owners newly reaching a judge are the owners who had been defaulting.

(c) alone is the hardest to read and the most decisive if it could be read. Contest rates are the residual of a decision nobody observes, and no dataset anywhere separates an owner who chose not to fight from an owner who never received the notice. A rise could mean better access, or it could mean the takings simply got larger. The three have to be read together, and (b) has to be read against the case mix of the year it came from.

What is it connected to?

Fills with researchcriminal fines, fees and bail as municipal revenue, police budgeting and the appropriations process generally, the composition of drug enforcement, and forfeiture regimes in other countries. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• How many people lost property, or how many properties were taken. The national account is denominated in dollars from end to end — USD 82 billion since 2000, roughly USD 300 million a year in state revenue — and the chapter that reports the totals contains no count. Meanwhile the requests made by the agencies are counted to the unit: 707,735 decided and 652,903 granted across the same span. What the government counts is who asked, not who lost.
• What the USD 1,678 figure actually is. The executive summary calls it a median cash forfeiture amount and the chapter on why owners fail to contest calls it an average. One number, two different statistics, one report. Nothing here picks between them, because the source does not.
• What share of any budget comes from forfeiture. This is the single number that would make the incentive measurable, and no jurisdiction publishes it routinely. It survives only as a survey statistic reported in a Supreme Court dissent, which means the claim that agencies depend on forfeiture can be asserted and denied indefinitely without either side being checkable.
• Who the owners are. Race, income, age and residence of the people whose property was taken are absent from every state dataset described. The only demographic evidence in the report is indirect — out-of-state residence in Kansas and Nebraska highway seizures, and one Philadelphia survey whose primary was not opened.
• Whether the reforms bind. No state that reformed publishes a before-and-after series on the federal participation of its own agencies, so the comparison has to be constructed from outside. A 2016 Nebraska reform contained a loophole that let most civil forfeitures continue, and that was discovered afterward by an outside organization rather than reported by the state that passed the law.
• Where the money goes after it arrives. No more than 12 percent of proceeds reached community programs and no more than 2 percent reached victim compensation across the 15 states with data, and under 1 percent of federal seized assets led to victim restitution. On transparency grading, 22 states earned a failing mark for accounting for fund spending and 30 earned one for financial audits.
• How to check the federal half. Treasury does not publish its forfeiture database online at all, and the DOJ tracking system is obtainable only through records requests and carries acknowledged reporting lags. The authoritative account of USD 57.4 billion in federal takings is a reconstruction by an advocacy organization, and no public counterpart exists against which that reconstruction could be checked. This round is consistent with that — justice.gov, treasury.gov, the Congressional Research Service and census.gov all refused the request.
• Whether the program is getting better or worse. The fourth national edition does not compare its own findings with the 2010, 2015 or 2020 editions. It sets no target, and neither does any statute or agency, so there is nothing against which the 2026 figures could be scored.

See the evidence

Four primary legal documents were read directly and everything else rests on one organization. The two Supreme Court opinions and the two statutes came from the Cornell Legal Information Institute and are the only non-advocacy sources in this file; the fund language at 28 U.S.C. 524(c), the Culley holding, the Timbs holding and the CAFRA deadlines are grounded there. Every quantity — the USD 82 billion, the 71 percent, the USD 1,678, the equitable sharing counts, the two state natural experiments — comes from a single advocacy organization, which built its federal series from records requests because the federal government does not publish the counterpart. That is the central limitation of this dossier and it is also the finding: the government publishes no dataset against which the reconstruction could be checked. Nothing in this document is quoted, because the research round could not re-verify exact wording against the pages. Second-hand material is labelled where it appears — the comparison of administrative against civil-judicial outcomes is the one finding reported at one remove that enters block 8, and it is graded low, while the Philadelphia survey is named in block 12 as a source whose primary was not opened. Where the source disagrees with itself the disagreement is left visible rather than resolved: the same USD 1,678 is called a median in one chapter and an average in another, the federal median currency forfeiture appears as USD 13,695 in one place and USD 13,980 in another for the same window, the grade counts sum to 42 of 51 jurisdictions with nine unenumerated, and the publication date reads 2026-03-24 on the landing page while the PDF sits at a path dated January. The date used throughout is 2026-03-24. This is a Path A output (research-based definition), so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 27,000–390,000 As of data through 2023

Sensitivity The unit is a property-taking event in a single year, not a person. No source anywhere publishes a count of people who lost property or of properties taken, and the entire national account is denominated in dollars, so the only count that exists is on the agency side of the transaction. The width of the band is not sampling noise but the unknown shape of the value distribution, which nobody publishes: the floor counts only the federal equitable sharing slice, and the ceiling divides dollar totals by medians, which always overstates a count in a right-skewed distribution. Events map to people in both directions at once, since one person can lose several properties, several people can claim one property, several agencies can file requests against a single seizure, and one request can cover several assets. In the other direction the band is a floor within a floor: property seized and later returned appears in no count, households affected by the loss of a single vehicle are not counted, owners deterred from contesting appear only as the residual in default rates that cannot separate them from owners who never received notice, and roughly one-third of the state datapoints across 24 years are missing or unknown. Narrowing the band requires an extrapolation assumption, and any such assumption is known to be wrong in both directions at once, because the state slice has a median under USD 1,678 while the federal slice contains billion-dollar corporate matters.

Regional breakdown State-level counts cannot be derived from anything opened this round. Reported values are dollar totals rather than property or person counts, three states report only seizures and never the forfeited end, roughly one-third of the state datapoints across 24 years are missing or unknown, and no United States or state population figure was sourced this round because census.gov refused the request and the remaining interface required a key. Apportioning a national figure by population share is not permitted and would be meaningless here in any case, since both the agency stake and the reporting regime differ by jurisdiction.