BRENNEN

Industries / Government, education and community / Government / Corrections

Probation and parole

Probation is supervision in the community that a court orders in place of prison, and parole is supervised release from prison before the end of a sentence; officers meet the people they supervise, help with housing and work, check the conditions set for them and report to courts and parole boards.

7

Milestones achieved, from 1841 to 2021.

2

To watch for: announced, not yet confirmed outside.

3

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1841

    A shoemaker stands bail

    In 1841 John Augustus, a Boston shoemaker aged 57, asked a judge to put off sentencing a man convicted of drunkenness for three weeks, paid his bail and supervised him. When the man returned sober, the judge fined him one cent and costs instead of sending him to the House of Correction. Augustus went on to screen the people he helped, keep a record of every case, find them work and housing, and give the practice its name, probation. In the same year Matthew Davenport Hill, a judge in Birmingham, England, began similar work. In 1878 Massachusetts passed the first state probation law, requiring Boston's mayor to appoint an officer to investigate cases and recommend probation to the courts. Charles Lindner, writing in the US courts' journal Federal Probation, notes that some historians, such as Samuel Walker, think Augustus's role has been exaggerated, since judges had long found ways to let people stay in the community under conditions.

    uscourts.govuscourts.gov

    Credit

    John Augustus; Matthew Davenport Hill; the Commonwealth of Massachusetts

  2. 1853

    Early release, no supervision

    From 1801 Britain's colony of New South Wales gave convicts of good conduct a ticket of leave, a permit to work for themselves, and in the 1840s Captain Alexander Maconochie ran the Norfolk Island penal colony on a marks system, in which good conduct earned better conditions and in time release. Walter Crofton later built a system of graded stages in Ireland's prisons that influenced American parole. England's Penal Servitude Act 1853 let prison authorities release people on license before the end of their sentence, but made no provision for supervising them afterwards. In 1863 a wave of violent robberies was widely blamed in Parliament on men recently released on license, and a Royal Commission that year recommended a formal marks system.

    LessonEarly release with no one checking on people afterwards lost public confidence within ten years.

    radar.brookes.ac.uk

    Credit

    Alexander Maconochie; Walter Crofton; history as told by Thomas Guiney (2018)

  3. 1939

    Volunteers supervise probation in Japan

    Japan's probation system grew out of volunteer work. In 1888 a group led by Meizen Kimpara and Kyoichiro Kawamura founded a society in Shizuoka to help people released from prison find homes and work, and Buddhist and Christian groups followed. The Judicial Rehabilitation Services Act of 1939 brought this work into the national system. Today about 47,000 volunteer probation officers, called hogoshi, are commissioned by the Minister of Justice but not paid, and work alongside salaried probation officers. In March 2021 a world congress in Kyoto adopted the Kyoto Hogoshi Declaration to spread the model to other countries.

    moj.go.jpmoj.go.jpmoj.go.jp

    Credit

    Meizen Kimpara and Kyoichiro Kawamura; Japan's Ministry of Justice

  4. 1983

    Ankle monitors begin

    In 1983 Judge Jack Love of Albuquerque, New Mexico, a former federal public defender, persuaded Michael Goss, a computer salesman, to build a system to monitor five people at home instead of in jail. A second system followed in Key Largo, Florida, and by 1987 21 US states reportedly had electronic monitoring programs covering more than 900 people. Ralph Schwitzgebel at Harvard had designed a prototype in 1964 and patented it, but it was too costly to spread. By the mid-2000s about a fifth of community supervision in the United States, and of people starting supervision in England and Wales, involved electronic monitoring, according to Ralph and Robert Gable.

    LessonThe prototype's designers meant it to reward progress, but it spread as a way to enforce conditions more strictly, as they themselves noted in 2005.

    uscourts.gov

    Credit

    Jack Love and Michael Goss

  5. 2016

    Risk scores put to test

    Risk assessment tools score how likely a person is to be arrested again, and courts, probation services and parole boards use them. In May 2016 ProPublica analyzed COMPAS, a tool sold by Northpointe, for more than 7,000 people arrested in Broward County, Florida, in 2013 and 2014. Of those it scored as likely to reoffend, 61% were arrested again within two years, and Black defendants who were not rearrested had been labeled higher risk 44.9% of the time, against 23.5% for white defendants. Northpointe disputed the analysis. Alexandra Chouldechova showed the same year that a score meeting one common test of fairness will produce unequal error rates when rearrest rates differ between groups. In 2018 Julia Dressel and Hany Farid found COMPAS no more accurate than people with little criminal justice experience, and a simple model using two inputs nearly matched its 137.

    LessonFairness has more than one definition, and when groups' rates differ a single score cannot meet them all.

    propublica.orgarxiv.orgebi.ac.uk

    Credit

    ProPublica (Julia Angwin, Jeff Larson, Surya Mattu and Lauren Kirchner); Alexandra Chouldechova; Julia Dressel and Hany Farid

  6. 2016

    Swift sanctions fail elsewhere

    In October 2004 Judge Steven Alm in Honolulu began HOPE, Hawaii's Opportunity Probation with Enforcement: people on probation were drug-tested at random, and every violation brought a short jail stay at once. In a randomized trial of 493 people begun in 2007, Angela Hawken and Mark Kleiman found that after a year 21% of those on HOPE had been arrested again, against 47% on ordinary probation, and that 13% of their drug tests were positive, against 46%. A replication in four other places, funded by the National Institute of Justice and published in 2016, assigned just over 1,500 people at random and found no significant drop in arrests, except drug arrests at one site; two sites revoked probation sooner, and the authors judged that savings were unlikely.

    LessonA striking result in one court did not carry over to four others, so one trial was not enough to spread a program.

    ojp.govnij.ojp.gov

    Credit

    Judge Steven Alm; evaluated by Angela Hawken and Mark Kleiman; replication funded by the National Institute of Justice

  7. 2021

    Probation returns to public hands

    In 2013 the Ministry of Justice dissolved the 35 probation trusts of England and Wales, and from February 2015 eight mainly private suppliers ran 21 Community Rehabilitation Companies that supervised people judged low or medium risk, while a public National Probation Service kept the higher-risk cases. The National Audit Office found in March 2019 that the reforms had achieved "poor value for money": the ministry would pay at least £467 million more than the original contracts required, and while fewer people reoffended, those who did committed more offenses each on average. The government had announced in July 2018 that the contracts would end early, and from 26 June 2021 one public probation service took over supervision of everyone on a community order or license.

    LessonSplitting probation between public and private providers by risk level was reversed within six years.

    nao.org.ukgov.uk

    Credit

    UK Ministry of Justice; reviewed by the National Audit Office

Watch for

  1. AI notes for probation officers

    The probation service of England and Wales uses Justice Transcribe, an AI tool that transcribes and summarizes supervision meetings. The Ministry of Justice says it has been rolled out to more than 12,000 probation officers and summarized more than 1.6 million meetings from 7 October 2025 to 14 September 2026. Its figure of about 266,667 hours saved rests on an assumed 10 minutes per meeting, which it calls illustrative, and it has published no measure of how accurate the summaries are.

    Achieved whenan outside evaluation compares a sample of summaries with the recordings and publishes the rate of errors and omissions.

    gov.ukgov.uk

    Claimed by

    UK Ministry of Justice, September 2026

  2. Face checks for remote check-ins

    The same ministry's one-year update on its AI plan lists Online Probation Check-ins, a platform that uses facial recognition to confirm the identity of people on probation when they report online instead of in person.

    Achieved whenan outside party publishes how often the face match wrongly rejects or wrongly accepts, broken down by age, sex and skin tone.

    gov.uk

    Claimed by

    UK Ministry of Justice, September 2026

Open

  1. 01

    Scripted supervision meetings test

    A public test of AI note-takers on recorded, made-up probation supervision meetings. Actors play an officer and a person on probation from written scripts, and each script holds facts a record must not miss or invent, such as a missed appointment, a change of address, a new job or a disclosure that raises concern. Each tool's summary is graded line by line against the script.

    The brief
    Credit · unclaimed
  2. 02

    Conditions in plain words

    Plain-language versions of the standard conditions of probation from every US state, and later from other countries, each checked by a lawyer to mean the same thing, and a test of whether readers understand the original or the rewrite better. The readers are adults recruited online, some of them screened for a low reading level; no one on probation is tested.

    The brief
    Credit · unclaimed
  3. 03

    A world probation count

    A dated table of how many people are on probation, parole or other supervision in the community in every country that publishes a figure, with the rate per 100,000 people, the definition each country uses and a link to the official source, updated once a year. It does for supervision in the community what the World Prison Brief does for prisons, and it holds national totals only.

    The brief
    Credit · unclaimed

Open 01 · the brief

Scripted supervision meetings test

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public test of AI note-takers on recorded, made-up probation supervision meetings. Actors play an officer and a person on probation from written scripts, and each script holds facts a record must not miss or invent, such as a missed appointment, a change of address, a new job or a disclosure that raises concern. Each tool's summary is graded line by line against the script.

Why it is still open

Justice Transcribe has summarized more than 1.6 million meetings, but the Ministry of Justice publishes usage counts and an assumed time saving, not accuracy. The Ada Lovelace Institute's February 2026 study of AI transcription in social work, based on interviews with 39 social workers in 17 local authorities, found that councils' evaluations focused on efficiency and that risks such as invented content "aren't being fully assessed or mitigated"; it did not measure error rates. Recidiviz, which builds software for US corrections agencies, said in June 2026 that it grades summaries from its own meeting tool as good, partial or bad, a bad one being a summary that could lead to the wrong response by an agency, and reduces human review only when error rates pass set thresholds; it publishes neither the rates nor a test set. No public test set of supervision meetings was found.

gov.uklocalgovernmentlawyer.co.ukrecidiviz.org

Built with what exists

Scripts written with former probation officers; recordings by paid actors with a range of accents; speech-to-text models such as ElevenLabs Scribe v3 (see the Note "Fewer word errors") and OpenAI's Whisper; summarizing models such as Claude, Gemini and GPT; a grading sheet agreed before the test.

  1. First step

    Write ten scripts with 15 must-record facts each, record two of them, and run both through two tools.

  2. Done looks like

    200 scripted meetings of 20 to 40 minutes, each run through at least four tools, with the share of scripted facts missing from each summary and the number of statements that were never said, graded blind by two people; scripts, audio and scores published.

Open 02 · the brief

Conditions in plain words

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

Plain-language versions of the standard conditions of probation from every US state, and later from other countries, each checked by a lawyer to mean the same thing, and a test of whether readers understand the original or the rewrite better. The readers are adults recruited online, some of them screened for a low reading level; no one on probation is tested.

Why it is still open

The Prison Policy Initiative's October 2024 report "One Size Fits None" collected standard conditions from 76 jurisdictions covering all 50 states and Washington, D.C., found an average of 12 standard conditions per person and called their language "complicated and unclear", but did not measure readability or understanding. A 2023 study of 74 prison handbooks from 49 states found all of them harder to read than their readers' likely literacy allowed; it did not cover probation conditions. No test of how well people understand conditions was found.

prisonpolicy.orgapi.semanticscholar.org

Built with what exists

The Prison Policy Initiative's published table of conditions; readability formulas such as Flesch-Kincaid and SMOG; a language model such as Claude or Gemini to draft rewrites; an online research panel such as Prolific for readers; a lawyer to check each rewrite.

  1. First step

    Take the conditions of five states from the report's table, draft rewrites, and write three comprehension questions for each condition.

  2. Done looks like

    Rewrites for all 76 jurisdictions; 1,000 readers each shown the original or the rewrite at random; for each condition, the share who answer correctly what it requires, with a lawyer confirming that no rewrite changes a legal duty; all materials and results public.

Open 03 · the brief

A world probation count

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A dated table of how many people are on probation, parole or other supervision in the community in every country that publishes a figure, with the rate per 100,000 people, the definition each country uses and a link to the official source, updated once a year. It does for supervision in the community what the World Prison Brief does for prisons, and it holds national totals only.

Why it is still open

The World Prison Brief counts people in prison, not people supervised in the community. The Council of Europe's SPACE II statistics cover Europe only, and 13 of the 52 rows in its table for 31 January 2025 have no total. The Global Community Corrections Initiative, run from the University of Massachusetts Lowell, published a table in 2020 for the 50 countries with the most people in prison, but its figures for supervision in the community date from 2010 to 2020, and for many countries it marks the data as not available. Searches found no newer world count.

globcci.orgwp.unil.ch

Built with what exists

National statistical yearbooks and the annual reports of corrections and probation agencies; the GLOBCCI and SPACE II tables as starting points; a low-cost model that reads long reports to find the figures, such as Muse Ember (see the Note "The jobs agents repeat"), with every figure checked by a person; a spreadsheet published as CSV.

  1. First step

    Fill the table for the 20 countries with the most people in prison, starting from the GLOBCCI table, and record which of them have published a newer figure.

  2. Done looks like

    Figures for at least 100 countries, each with its reference date, definition and source; 20% of rows rechecked by a second person with at least 95% agreement; a world total published with the countries it leaves out.