Industries / Government, education and community / Government / Corrections
Health care inside
Health care inside is the medical, dental and mental health care given to people while they are held, including the control of infections that spread in crowded buildings and preparation for the risky weeks after release.
7
Milestones achieved, from 1976 to 2023.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1976
A right to medical care
On 30 November 1976 the US Supreme Court decided Estelle v. Gamble, brought by J.W. Gamble, who was hurt in 1973 when a bale of cotton fell on him during prison work in Texas. Writing for the Court, Justice Marshall held that "deliberate indifference to serious medical needs of prisoners" is cruel and unusual punishment, which the Eighth Amendment forbids. Gamble's claim against the prison's medical director failed: doctors had seen him 17 times in three months, so any fault was at most malpractice, a matter for state courts.
LessonThe ruling set a floor, not a standard of care; poor care that is not deliberate indifference is left to malpractice law.
Credit
US Supreme Court
1999
Standard TB treatment fails
At a tuberculosis referral prison in the Kemerovo region of Siberia, 35% of treatments failed between June 1996 and March 1997, even though nurses watched every dose (DOTS, directly observed treatment, short course) and used the World Health Organization's standard re-treatment drugs. Among 164 patients tested from December 1997 to March 1998, 22.6% had TB that resisted the two main drugs, isoniazid and rifampin, from the start: multidrug-resistant TB.
LessonWhere drug-resistant TB is common, watching every dose of the standard drugs is not enough; prisons need drug-resistance testing and regimens that match it.
Credit
Kimerling ME, Kluge H and colleagues, in the International Journal of Tuberculosis and Lung Disease
2011
Naloxone handed out at release
Naloxone reverses an opioid overdose. In January 2011 Scotland's National Naloxone Programme began giving brief training and naloxone kits to people at risk of overdose, in the community and in all prisons. Comparing 2006–10 with 2011–13, the share of opioid deaths that happened within four weeks of leaving prison fell from 9.8% (193 of 1,970) to 6.3% (76 of 1,212), a 36% reduction and about 42 fewer deaths; 12,000 kits were issued in 2011–13 at a prescription cost of £225,000.
Credit
Scotland's National Naloxone Programme; evaluated by S. M. Bird and colleagues in Addiction (2016)
2011
Crowding ruled a health risk
California's prisons were designed for just under 80,000 people and held almost twice that. In a case filed in 2001 the state had conceded that its prison medical care violated the Eighth Amendment, and a court-appointed Receiver later reported that crowding kept care from improving. On 23 May 2011, in Brown v. Plata, the US Supreme Court upheld 5–4 an order to cut the population to 137.5% of design capacity within two years.
LessonThe courts concluded that medical and mental health care could not be fixed while the buildings held about twice the people they were designed for.
Credit
US Supreme Court, opinion by Justice Kennedy
2016
Addiction medicine behind bars
In July 2016 the Rhode Island Department of Corrections began screening people for opioid use disorder and offering all three approved medicines, methadone, buprenorphine and naltrexone; all its sites were running the program by January 2017. In the first half of 2017, 9 of 157 people who died of overdose in the state had recently left prison or jail, against 26 of 179 in the first half of 2016, which the authors report as a 60.5% fall in deaths in this group; overdose deaths statewide fell 12.3%.
Credit
Rhode Island Department of Corrections; measured by Green TC, Rich JD and colleagues in JAMA Psychiatry (2018)
2019
Hepatitis C treated at scale
In the SToP-C study in four Australian prisons, more than 3,600 people were tested for hepatitis C over five years, and those infected were offered direct-acting antivirals, pills taken for 12 weeks in this study. The researchers reported in 2021 that broad testing and treatment halved new infections inside. A 2024 analysis found that 96% of people with a documented end of treatment were cured, but only 44% of all who started, because release or transfer to another prison cut off treatment or follow-up for nearly all of those whose results were missing.
LessonPrison treatment works as well as treatment outside only if release and transfers are planned into it.
Credit
Kirby Institute, UNSW Sydney, led by Gregory Dore and Andrew Lloyd
2023
AI reads prison X-rays
Researchers screened 2,075 men in three Brazilian prisons for tuberculosis between October 2017 and December 2019, with chest X-rays in a mobile unit and sputum tests; 259 (12.5%) had TB. They then scored three AI programs that read chest X-rays: CAD4TB, Lunit and qXR. All three performed similarly overall, but at 90% sensitivity only Lunit and qXR met the World Health Organization's target for a triage test (specificity 84% and 74%), and none met it for people who had had TB before.
LessonGood overall accuracy hid weaker results in one group, so a screening tool has to be checked in each group it will screen.
Credit
Soares TR and colleagues, Lancet Regional Health – Americas (2023)
Watch for
Injections before release
North Carolina's Department of Adult Correction gives people a few days of oral buprenorphine and then a month-long buprenorphine injection (Brixadi) three to five days before release, so they leave prison with treatment that lasts until they reach a community provider. The pilot began in 2021 in two prisons and spread in 2025 to 24 facilities, about half the state's prisons; the department's chief medical officer, Arthur Campbell, says more than 1,000 people received it in the past year. Evan Ashkin, who directs the state's Formerly Incarcerated Transition Program, says it is "making a tremendous difference", but no outcome figures have been published.
Achieved whenoutside researchers publish overdose deaths in the weeks after release for people who received the injection, compared with similar people who did not.
Claimed by
North Carolina Department of Adult Correction and NC FIT, December 2025
Open
01
Prison health contracts compared
An open dataset of the contracts that US state prison systems and large county jails sign with private health care companies, each coded the same way: how the company is paid (per person per day, a fixed sum, or costs plus a fee), the price, caps on hospital bills, required staffing, response times, penalties and reporting duties. It uses public contracts and procurement records only, never information about patients.
The briefCredit · unclaimed02
A prison health question set
A public set of questions with reference answers drawn from published prison health guidance, such as which people should be screened for tuberculosis on arrival, or what a prison must do in a heat wave, used to test whether AI assistants give correct answers before prison staff rely on them. Each answer cites the passage it comes from, and two licensed clinicians grade each model's answers. No patient information and no treatment are involved.
The briefCredit · unclaimed
Open 01 · the brief
Prison health contracts compared
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open dataset of the contracts that US state prison systems and large county jails sign with private health care companies, each coded the same way: how the company is paid (per person per day, a fixed sum, or costs plus a fee), the price, caps on hospital bills, required staffing, response times, penalties and reporting duties. It uses public contracts and procurement records only, never information about patients.
Why it is still open
The Pew Charitable Trusts sorted states by how they delivered prison health care in fiscal 2015 (17 mainly by their own staff, 20 mainly by contractors, 8 a mix, 4 with a state medical school) and found that 19 of the 28 that used contractors for most or half of care paid a fixed amount per person, but it did not publish the contracts' prices, staffing terms or penalties. The American Friends Service Committee counted 36 states that contract out at least part of prison health care, 24 of them all of it, but published no contract terms. Reuters surveyed 523 jails in 2020, found death rates 18% to 58% higher in jails using the five leading contractors than in publicly run jails from 2016 to 2018, and reported that contracts "sometimes lack quantified standards for care", such as staffing requirements; it did not publish a coded set of contracts. Searches found no public dataset of contract terms.
Built with what exists
State procurement portals and public records requests; pdftotext and Tesseract for scanned contracts; a small, low-cost model for repeated extraction, such as Meta's Muse Ember (see the Note "The jobs agents repeat"); a shared coding sheet with the page reference for every value.
First step
Download the current health care contracts from five states' procurement sites and draft the coding sheet.
Done looks like
Current contracts from at least 30 of the 36 contracting states and 100 county jails, coded and published; on 300 values checked by hand against the contracts, at least 98% match.
Open 02 · the brief
A prison health question set
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of questions with reference answers drawn from published prison health guidance, such as which people should be screened for tuberculosis on arrival, or what a prison must do in a heat wave, used to test whether AI assistants give correct answers before prison staff rely on them. Each answer cites the passage it comes from, and two licensed clinicians grade each model's answers. No patient information and no treatment are involved.
Why it is still open
The US Federal Bureau of Prisons publishes 37 clinical guidance documents "for informational purposes", among them tuberculosis (2020), heat-related illness (2017) and opioid use disorder (2024). Searches of Hugging Face datasets and arXiv on 18 November 2026 found no question set built from prison health guidance.
Built with what exists
The Bureau's guidance PDFs and pdftotext; several AI models, including small open-weight models a prison system could run on its own servers; a grading sheet; two clinicians paid as graders.
First step
Write 50 questions with answers and page references from the Bureau's tuberculosis and heat-related illness guidance.
Done looks like
500 questions from at least 15 guidance documents, at least five models tested, and the share of answers both graders rate as wrong or unsafe published for each model with the questions and answers.