KOHEN AVOCATS · PARIS, FRANCE
When Does Elizabeth Holmes Get Out of Jail? Federal Good-Time Credit, Halfway House and Appeal Checks

Hassan Kohen is a Paris-based lawyer admitted to the Paris Bar (Avocat au Barreau de Paris). Initial strategic analysis: €80 incl. VAT, personal reply within 24 hours.

100% confidential · Legal professional privilege · No obligation

Article generated by artificial intelligence, through a process designed and supervised by the firm

Sources: U.S. court opinions and official U.S. government publications, read for this article

Maître Hassan KOHEN, avocat au Barreau de Paris
Hassan KOHEN
Attorney at the Paris Bar

When Does Elizabeth Holmes Get Out of Jail? Federal Good-Time Credit, Halfway House and Appeal Checks

Speak to a lawyer 06 89 11 34 45

Search interest in the question of when Elizabeth Holmes will leave federal prison has surged again, with readers asking in plain terms when she gets out of jail and why her projected date keeps moving. Press reports describing Bureau of Prisons transfer paperwork put her current projected release in February 2030 and describe a move to a Texas halfway house next summer, after earlier projections that stretched into 2032 and beyond. Those shifting dates confuse many readers: a federal judge imposed eleven years and three months, the court of appeals affirmed, and yet the calendar keeps changing without any new trial. The discussion that follows is general in scope: it describes checks useful in comparable situations and makes no findings about any individual’s case beyond the published court decisions it cites. Federal law answers the puzzle through several distinct mechanisms that stack on top of one another. Good-conduct credit shortens the calendar automatically, earned credits under the First Step Act can move a prisoner into prerelease custody sooner, a halfway-house placement covers the final months, and only narrow statutory paths allow a judge to revisit the sentence itself. This guide walks through each mechanism in the order a lawyer would check them, using the published decisions in her case and three comparable federal rulings that explain the math courts actually apply.

I. What sentence was imposed, and why does the release date keep moving?

A. The four counts of conviction and the affirmed 135-month sentence

In late 2021, after a trial of nearly four months, a federal jury in the Northern District of California acquitted Elizabeth Holmes on the patient-related counts, hung on three investor wire-fraud counts, and convicted her of four investor-related counts: conspiracy to commit wire fraud against investors and three substantive wire-fraud counts tied to three investor victims. Wire fraud is the federal charge that fit the case: 18 U.S.C. section 1343 punishes anyone who, having devised a scheme to defraud or to obtain money by false pretenses, transmits writings, signals, or sounds by wire in interstate commerce to execute it, with up to twenty years of imprisonment, or up to thirty where a financial institution is affected. At sentencing, the district court found investor losses of 120 million dollars, producing a 24-level enhancement under Guidelines section 2B1.1(b)(1) and a recommended range of 135 to 168 months, and imposed 135 months of imprisonment with a restitution order covering the investors’ full investment amounts.

The restitution piece of the judgment deserves its own warning because readers often mistake a large dollar figure for leverage over the release date. The Ninth Circuit explained that the investors’ inability to resell their shares justified awarding the full value of each investment as restitution, to make the victims whole, while noting that loss calculation for sentencing serves a different punitive purpose and does not necessarily require the same offset. It then affirmed the district court’s restitution order in its entirety. The application is straightforward and frequently misunderstood: restitution is a money judgment that survives independently of the prison calendar, payment plans and collection efforts do not move a projected release date, and a prisoner who finishes the custodial term still owes whatever balance remains. For anyone comparing this case to another fraud prosecution, the loss amount drives the Guidelines range at sentencing, as the 120-million-dollar finding and the 24-level enhancement show, but once the sentence is final the loss figure drops out of the release-date math entirely.

On direct appeal, the United States Court of Appeals for the Ninth Circuit affirmed everything that mattered to the calendar. In United States v. Elizabeth Holmes, 129 F.4th 636 (9th Cir. Feb. 24, 2025), the court rejected the challenges to the convictions, the 135-month sentence, and the restitution order, affirming the district court’s judgment in its entirety, and an amended opinion issued on December 22, 2025 left that result in place while denying rehearing. The opinion issued under Ninth Circuit docket number 22-10312. For a reader checking whether some appeal might still undo the sentence, the useful application is direct: the affirmance means the 135-month term is the lawful starting point for every later calculation, and no pending direct appeal suspends it. The affirmance decides the validity of the convictions, the 135-month term and restitution; it does not compute good-time or earned-time credits or set the Bureau release date, which remain governed by sections 3624 and 3632. That finality point was promptly treated as governing circuit law: in United States v. Greene, No. 23-4097 (9th Cir. May 23, 2025), the Court of Appeals for the Ninth Circuit recited the plain-error test from Holmes when reversing a Guidelines calculation built on a carjacking pseudo-count, which confirms that later panels cite Holmes as the circuit’s statement of that standard of review. A sentence that has survived appeal in this way can still produce a moving release date, because the release date is computed by the Bureau of Prisons under separate statutes, not fixed on sentencing day.

B. Why a Bureau of Prisons date moves without any new court order

The release date on a federal inmate record is a projection, not a second sentence, and Congress assigned its computation to the Bureau of Prisons. The principal engine is good-conduct credit under 18 U.S.C. section 3624(b), which lets a prisoner serving more than a year earn credit toward service of the sentence, beyond time served, of up to 54 days for each year of the sentence imposed by the court, subject to the Bureau’s determination that the prisoner has displayed exemplary compliance with institutional rules. The Supreme Court of the United States endorsed the Bureau’s arithmetic in Barber v. Thomas, 560 U.S. 474 (June 7, 2010), concluding that the Bureau’s method reflects the most natural reading of the statute and rejecting the prisoners’ challenge, which for readers means the credit accrues on time actually served as each year is completed rather than as a flat discount off the imposed term. Congress later clarified the rate in the First Step Act of 2018, substituting credit of up to 54 days for each year of the prisoner’s sentence imposed by the court, and the Bureau’s regulation restates the practice plainly: 28 C.F.R. section 523.20 provides that the Bureau will award inmates up to 54 days of good-conduct credit for each year of the sentence imposed by the court. Applied to a 135-month term, that credit alone removes well over a year from the calendar, which is why an eleven-year-and-three-month sentence imposed in 2022 can project to a 2030 release even before any other mechanism is considered.

A rough illustration shows how powerful the good-time line of the worksheet is. Eleven years and three months is 135 months, or eleven and one-quarter years. At up to 54 days per year of the sentence imposed, the maximum good-conduct award is on the order of six hundred days, roughly twenty months, before a single earned-time credit or halfway-house day enters the picture. That illustration is deliberately approximate: the Bureau prorates partial years, denies or restores credit based on disciplinary findings during each year, and applies the First Step Act’s clarified text prospectively within its regulation, so no two prisoners with the same imposed term necessarily share a projected date. Thieme’s proration holding matters exactly here, because the final partial year of a 135-month term generates only a fraction of the 54-day figure, and small differences in that fraction explain why locator dates drift by weeks between updates. Readers doing their own estimate should treat the statutory maximum as a ceiling, subtract nothing for programming credits they have not verified as earned and applied, and remember that only the Bureau’s computation sheet states the official result.

Courts continue to enforce that arithmetic against Bureau miscalculations, and a recent appellate decision shows how close the disputes get. In Christopher Thieme v. Warden Fort Dix FCI, No. 23-1697 (3d Cir. Oct. 8, 2025), the Third Circuit walked through Barber’s reasoning that the statute created a retrospective award rather than a prospective entitlement, held that good conduct time credits accrue at the rate of 54 days per year with partial years credited on a prorated basis, and affirmed the district court’s judgment applying that math. The direct application for anyone reading an inmate locator page is practical: when a projected date shifts by weeks or months with no court filing, the first hypothesis is a good-time recalculation or the crediting of a partial final year, and the remedy for a genuine error is a habeas petition challenging the execution of the sentence, the same vehicle Thieme used. That vehicle matters because later sections of this guide turn on it: both good-time disputes and earned-credit disputes reach federal court, if at all, through petitions under 28 U.S.C. section 2241, which authorizes the writ for federal prisoners challenging how their sentence is being carried out rather than the validity of the conviction itself.

II. Which federal rules can shorten custody further, and which cannot

A. First Step Act credits and the halfway-house door

A second and distinct shortening mechanism comes from the First Step Act’s earned-time program. Under 18 U.S.C. section 3632(d), eligible prisoners who complete Bureau-designated evidence-based recidivism-reduction programming and productive activities earn time credits that the Bureau may apply toward prerelease custody or an earlier start of supervised release, as implemented by 28 C.F.R. section 523.44, which channels those credits toward prerelease custody or supervised release for qualifying prisoners. These credits differ from good-conduct time: they must be earned through programming, they depend on the prisoner’s assessed recidivism risk level, and certain offenses and criminal histories are excluded by statute. In Gonzalez v. Herrera, No. 24-2371 (9th Cir. Aug. 19, 2025), the Court of Appeals for the Ninth Circuit held that the district court misapplied section 3632(d)(4)(C) in a habeas case brought by a prisoner seeking recalculation of his earned credits, reversed the dismissal of the petition filed under 28 U.S.C. section 2241, and remanded with instructions to grant it in part, direct the government to recalculate the credits, and provide the recalculation to the probation officer. The application for readers is concrete: when transfer paperwork shows movement toward prerelease custody that good-time math alone cannot explain, earned-time credits are the next line on the worksheet to verify, and Gonzalez confirms that courts will order the Bureau to redo the calculation when the agency applies the wrong subparagraph.

The halfway-house transfer itself operates under a third provision that caps how much of the end of a sentence may be served in the community. 18 U.S.C. section 3624(c) directs the Bureau to ensure, to the extent practicable, that a prisoner serving a term of imprisonment spends a portion of the final months of that term, not to exceed 12 months, under conditions offering a reasonable opportunity to adjust to and prepare for reentry, expressly contemplating placement that may include a community correctional facility, alongside a separate home-confinement authority. Press reports describing the transfer paperwork for the prisoner discussed here mention approval for a 460-bed Austin facility with a projected release date in February 2030, which fits this framework: a residential reentry placement of several months at the end of the term, with residents leaving for work and programming but returning for curfew and testing, rather than a release from the sentence. Readers should keep the boundary clear. A halfway-house placement does not cut the sentence, does not reflect a judicial finding about the underlying convictions, and does not guarantee any particular total time inside, because the Bureau sets its length case by case within the twelve-month ceiling. Anyone checking a similar transfer should therefore ask three questions in order: how many good-time days have vested, how many earned-time credits have been applied toward prerelease custody, and how many of the final twelve months the Bureau has designated for community placement.

B. Sentence modification, finished appeals, and clemency requests

A federal judge generally cannot revisit a prison term once imposed, and the narrow statutory exceptions explain why most release-date movement happens at the Bureau rather than in court. 18 U.S.C. section 3582(c) provides that a court may not modify a term of imprisonment once imposed except in defined circumstances, including a motion by the Bureau’s Director or by the defendant after fully exhausting administrative appeals over the Bureau’s failure to bring such a motion or after thirty days have lapsed since the warden received the request. That exhaustion gate is strict and frequently dispositive: a letter to the judge, a press interview, or a change in public attention does not substitute for it. For the prisoner discussed here, the direct appeal has already run its course through affirmance and denial of rehearing, so section 3582(c) motions and habeas challenges to the execution of the sentence are the only judicial vehicles left, and each requires its own statutory predicate rather than a general plea that the sentence feels long. A further practical check applies to any talk of a reduced judicial role for old conduct: the court of appeals upheld the loss finding, the 24-level enhancement, and the bottom-of-the-range 135-month sentence together, which leaves little room for a later claim that the sentencing math itself was mistaken.

Choosing the right courthouse door is itself a check worth making explicit. A prisoner who believes the conviction or the imposed sentence is legally invalid generally proceeds by a motion attacking the sentence, while a prisoner who accepts the sentence but disputes how the Bureau is executing it, through good-time math, earned-time math, or prerelease placement, proceeds by a habeas petition under section 2241 in the district of confinement. Gonzalez and Thieme both reached the courts of appeals through that second door, and both decisions turned on execution questions rather than guilt or the length the judge selected. Filing in the wrong door costs months: courts routinely recharacterize or dismiss mislabeled petitions, limitations and second-or-successive rules police the attacking-sentence route far more tightly, and exhaustion of the Bureau’s administrative remedy program is expected before a judge reaches the merits of an execution claim. Anyone helping a federal prisoner document a release-date dispute should therefore sort papers into two piles from the start, judgment and sentencing papers in one, computation and programming records in the other, and match each grievance to the vehicle Congress assigned to it.

Executive clemency sits entirely outside that judicial framework, and readers should evaluate reports about it with the same discipline. Press accounts describe a commutation request submitted in early 2026 through the Justice Department’s pardon process, and a request alone changes nothing on the Bureau’s worksheet: credits continue to accrue or not under the statutes above while any request is pending, and only an actual grant alters the term. That distinction answers the most common confusion in the search queries. A headline about a halfway-house transfer reflects prerelease placement under section 3624(c), possibly accelerated by earned-time credits under section 3632. A headline about a judge cutting a sentence would have to point to a section 3582(c) order or an amended judgment, neither of which appears in the published appellate record through the December 2025 amended opinion. And a headline about a commutation would have to point to a signed grant, not a filing. Until one of those documents exists, the projected date remains what the transfer paperwork describes: a Bureau-computed calendar, not a judicial declaration that the punishment was wrong.

Finality also explains why later court activity in the same docket rarely helps a release-date argument. The December 2025 amended opinion in the Holmes appeal addressed the parties’ rehearing petitions, denied them, and left the affirmance intact, which is the ordinary end of direct review in the court of appeals. After that point, the vehicles narrow sharply: a petition for certiorari to the Supreme Court, which the Court grants in only a small fraction of cases and which does not stay the sentence by itself, or collateral and execution-phase challenges that each carry their own gatekeeping rules. None of those vehicles recalculates good-time or earned-time credit as a matter of course. A reader who sees a new docket entry and wonders whether the date will move again should therefore identify the document first: an amended opinion on rehearing leaves the calendar untouched, a Bureau recalculation memo moves it, and only a new judgment or a signed clemency grant changes the underlying term.

A last practical note concerns how to read the Bureau’s own public numbers while this computation runs. The inmate locator shows a projected release date that bundles every mechanism above into a single day, so a one-month shift between two visits to the page cannot by itself reveal whether good-time credit posted, earned-time credits were applied, or a prerelease placement was approved. The documents that distinguish those causes are the sentence computation sheet, the risk-and-needs assessment and programming records behind any earned-time award, and any written prerelease placement approval, all of which a prisoner can request through administrative channels and, if necessary, challenge through the exhaustion-and-habeas path described above. Press reports of a February 2030 date and a summer halfway-house transfer describe the output of that worksheet, not its reasoning, which is why this guide has worked through the worksheet line by line instead of treating any single reported date as the end of the story.

One more boundary keeps the whole picture honest. None of the credit mechanisms reviewed here revisits guilt, reweighs the trial evidence, or endorses any narrative about the underlying business. They answer a narrower administrative question that arises after every federal sentencing: given a final judgment, on which day does the law require the doors to open. Keeping that question separate from debates about the verdict is what lets each new headline slot into its proper line of the worksheet.

Conclusion

The answer to the reader’s question has two layers. The calendar layer is arithmetic: a 135-month affirmed sentence, reduced by good-conduct credit of up to 54 days per year, adjusted by any earned-time credits a court has ordered the Bureau to recalculate correctly, and finished through up to twelve months of prerelease placement that can include a halfway house. The legal layer is finality: convictions affirmed on direct appeal stand, the Bureau’s math is reviewable only through the narrow habeas channel for the execution of a sentence, and a judge may modify an imposed term only through the gated exceptions Congress wrote into section 3582(c). Anyone following a similar federal case should therefore read each new headline against those layers, asking which mechanism the underlying document actually invokes, and should treat a moving projected date as normal Bureau computation unless a fresh court order or signed grant says otherwise.

📄 Official circular