The Reform That Did Not Stick

It lasted 187 days.

On February 21, 2017, a new Attorney General rescinded it in a single paragraph, on the grounds that the memo had "impaired the Bureau's ability to meet the future needs of the federal correctional system." On January 26, 2021, a new President reinstated the policy by executive order. In January 2025, that order was revoked in a first-day batch alongside dozens of others, most of which nobody read.

Four official positions in nine years. Zero statutes.

This is not a story about prisons. Prisons are where it happened to occur. It is a story about a category error so popular it has gone invisible: reading the instrument a policy was written on as the reason it failed.

i · the half-life of a memo

The 2016 decision was, by the standards of executive action, unusually well-grounded. It followed an Inspector General review that compared contract prisons to comparable federal institutions and found more contraband, more assaults, more uses of force, and more lockdowns per capita — while honestly noting that the populations were not perfectly matched. The federal prison population was falling. The rationale was documented, the timing was defensible, and the moral case was clean.

None of it survived a change of pen.

The recursion is old and well documented. The Mexico City Policy governing foreign aid and abortion counseling was instituted by executive action in 1984, rescinded in 1993, reinstated in 2001, rescinded in 2009, reinstated and expanded in 2017, rescinded in 2021. Six reversals, four decades, one sentence of legislation: none. The Paris Agreement: in, out, in, out. DACA: created by memo in 2012, terminated by memo in 2017, litigated for years precisely because nobody could point to a statute. Keystone XL: permit, revoked permit, permit, revoked permit.

From which the obvious lesson gets drawn every time: memos are weak, statutes are strong, write it into law.

The obvious lesson is wrong, and it is worth being precise about why, because the correct version is more useful.

Statutes die. Glass-Steagall was repealed after sixty-six years. The Voting Rights Act's coverage formula — a statute Congress reauthorized five times, the last of them by a 98-0 Senate vote — was gutted by a court in 2013, and Congress has not restored it since. The individual mandate's penalty was zeroed out through budget reconciliation. Meanwhile Executive Order 9981, which desegregated the armed forces in 1948, is still standing seventy-eight years and fourteen presidents later, and nobody has ever seriously moved to rescind it. That one is a memo. It has outlived most of the statutes of its era.

So the paper is not the mechanism. The paper is the diagnostic.

A memo is what you write when you cannot get a statute. It is the instrument available to a coalition too weak to change the underlying arrangement — which is exactly why memo-reform correlates so tightly with reversal. The weakness is upstream of the document. By the time an administration is reaching for the Deputy Attorney General's letterhead, the informative fact has already occurred: nobody could assemble the votes, the appropriation, or the constituency. The memo is not the cause of death. It is the receipt.

EO 9981 held because the demand underneath it had genuinely moved — a war had just been fought, the manpower argument was finished, and the counter-coalition had nowhere left to stand. The Voting Rights Act formula fell because a statute's protective coalition had eroded to the point where a court could remove it and a legislature could not put it back. Same lesson from both directions. The document is downstream. The coalition is the load-bearing element.

ii · the market priced the election

On the day of the 2016 announcement, shares in Corrections Corporation of America fell about thirty-nine percent. GEO Group fell about forty percent — the largest single-day decline in its history as a public company. This is usually cited as evidence that the memo mattered.

It is evidence that the memo mattered. That is the interesting part. The market genuinely believed it, and repriced two companies by roughly two-fifths in an afternoon on the strength of one signature.

Then came November 8, 2016. By inauguration, both stocks had roughly doubled. CCA — which had rebranded itself CoreCivic that October, a move with the tonal quality of a defendant getting a haircut before trial — spent 2017 posting perfectly ordinary earnings.

Note what that sequence actually shows, because it is easy to overclaim here. The market did not foresee the memo's fragility in August; nobody prices a rescission that does not yet have an author. What it did was hold the reform at exactly the value of one election, and settle up the moment the election resolved. The worth of the policy, in the market's assessment, was the worth of a single coalition holding power for a specified term. That assessment was correct to the day.

There is a structural reason the industry could absorb the swing in either direction. Effective January 1, 2013, both major operators converted to real estate investment trusts — a tax structure built around distributing a reliable stream of income to shareholders, and one that cut their effective tax rates by thirty percent or more. That structure only works if the income stream is reliable. A company organized as a REIT around government detention contracts is not a company that will be dissuaded by a memo; it is a company whose entire capital structure is a standing bet on the durability of the underlying demand.

The bet was correct. The demand was never in question. Only the paperwork was.

iii · what the memo never touched

Here is the sentence that should have led every story in 2016 and led almost none of them.

The policy applied to thirteen contract prisons holding roughly twenty-two thousand people. That is about twelve percent of the federal prison population — and well under two percent of the roughly 1.5 million people held in American prisons at the time.

It did not apply to state prisons, which hold the overwhelming majority of incarcerated Americans and where the private contracting relationships are deepest and largest. It did not apply to immigration detention, which is run by a different department entirely, and where the large majority of beds were privately operated. When the 2021 executive order arrived, it carried the same limitation, by design: it directed the Justice Department not to renew contracts for privately operated criminal detention facilities. Immigration detention sits at Homeland Security. It was never in scope.

So the most-covered private-prison reform of the last decade governed under two percent of American incarceration and explicitly excluded the fastest-growing segment of the private detention market.

This is the standard shape of memo-reform: maximum symbolic surface, minimum structural contact. Not because the people writing it are cynics — the 2016 memo appears to have been written in good faith by people who had read the Inspector General's report and were appalled by it — but because the instrument a weak coalition can reach extends only as far as that coalition already governs. Executive action is bounded by executive jurisdiction. That boundary is invisible in the headline and decisive in the outcome.

iv · the twenty-two thousand

It is worth stopping on that number, because every paragraph above has been using it as a denominator.

Twenty-two thousand people were held in the thirteen facilities the memo covered. The Inspector General's finding — the one that made the memo defensible — was not an abstraction about contract compliance. It was that those facilities recorded more assaults, more uses of force, more contraband, and more lockdowns than comparable federal institutions. That is a report about what happened to specific people in specific buildings, converted here, as it was converted in most of the 2016 coverage, into evidence for an argument about governance.

The people in those buildings could not read a calendar and budget for a reversal. They could not diversify into an adjacent niche. Four official positions in nine years is, from a desk, a pattern; from inside, it is a decade in which the conditions of your confinement were a function of which coalition held the pen, and in which no stage of the argument — not the memo, not the rescission, not the executive order, not the revocation — was ever conducted with any of you in the room.

That is the part of the mechanism worth naming out loud. The parties who can read a calendar keep winning against the parties who cannot, and the paperwork keeps filing the result as a dispute about federal contracting. A settlement reached by excluding the party it acts upon is not a resolution. It is an arrangement among everyone who could afford a seat.

v · the money does not leave, it relocates

Corrections Corporation of America was founded in January 1983. Its first contract, awarded later that year, was to build and run an immigration detention facility in Houston for the Immigration and Naturalization Service; the doors opened in 1984. The industry was born in immigration detention.

So when the federal criminal contracts came under pressure, the industry went home. Immigration detention revenue grew. The operators diversified into residential reentry facilities, halfway houses, and electronic monitoring.

Every reform that targets a vendor rather than a demand produces this. Convict leasing in the American South was abolished state by state over decades — Alabama, the last holdout, ended it in 1928. The forced labor did not end. It moved to chain gangs and state-run prison farms, which in several states outlived the leasing system by a half-century. The vendor changed. The extraction continued. The abolition was real, and it was real in the way a name change is real.

Kill the contractor, and the function finds a new contractor. Kill the function, and the contractor has nothing to sell.

vi · what actually holds

There is a control group here, and it is instructive — though not for the reason it usually gets cited.

The Fair Sentencing Act of 2010 reduced the crack-versus-powder cocaine sentencing disparity from 100:1 to 18:1. The First Step Act of 2018 expanded earned time credits and made portions of the 2010 reform retroactive. Both have survived multiple administrations and changes of party control. The First Step Act passed 87-12 in the Senate and 358-36 in the House, and was signed by a President whose Attorney General had personally rescinded the private prison memo twenty-two months earlier.

The tempting reading is that they held because they are statutes. The better reading is that they are statutes because they could hold — because a coalition wide enough to produce those margins existed first, and legislation is the residue such a coalition leaves behind. The durability and the document are both outputs of the same upstream fact, and only one of them is visible in the coverage.

Which means the useful test is not "is it a law." The useful test is: what would reversing this cost, and who would have to pay it? A statute with a 98-0 coalition behind it is close to immovable; the same statute forty years on, its coalition dispersed, comes apart in a single 5-4 decision. An appropriation is durable in proportion to who shows up to fight for it. A facility that has been physically closed and sold is durable because concrete does not care about elections.

By that test the 2016 memo was correct on the merits and, structurally, a suggestion. It said: stop doing this. It did not say: this is now illegal, or there is no longer money for this, or the facilities have been closed and the contracts voided. It asked a bureaucracy embedded in a demand structure it did not control to please behave differently — which is moralizing with letterhead, and moralizing is what is left over when the leverage has run out.

vii · the vendor that sells the cure

Which brings us to the part of this story that ought to be a graduate seminar and currently gets a sentence in the trade press.

GEO Group owns BI Incorporated. BI has held the contract for the federal government's flagship alternative-to-detention program since 2004 — the Intensive Supervision Appearance Program, renewed in 2020 on a five-year award reported at up to $2.2 billion, monitoring on the order of a hundred and eighty thousand people through ankle bracelets, a mandatory check-in smartphone app, and a GPS smartwatch.

Now read that against the reform. The demand is do something about detention. The prescribed remedy is detain fewer people; supervise them in the community instead. The company that runs the detention facilities also runs the community supervision. Whichever way the reform goes, the invoice arrives at the same address.

That is not a hedge. A hedge is insurance against loss. This is a different animal: an industry that has made itself reform-proof by acquiring both sides of its own abolition. There is no coalition you can assemble, no statute you can pass, no appropriation you can cut that threatens a vendor positioned to be paid for the alternative to its own product. You cannot beat it by winning. Winning is one of its revenue lines.

And that is the real reason the 187 days did not matter, and would not have mattered at 187 weeks. The memo was aimed at a contractor. The contractor had already moved. The function — hold people, watch people, bill the government per person per day — was never on the table, because nobody in the fight was strong enough to put it there.

Which is precisely why the fight was being conducted on letterhead.