{
  "record": {
    "id": "fc726910-7daf-4fce-9d19-9b6ac7ea12ae",
    "code": "PAI-FC726910",
    "force": "Hertfordshire Constabulary",
    "country": "United Kingdom",
    "country_code": "uk",
    "is_country": true,
    "region": "Hertfordshire",
    "technology": "ADA (Anathem Digital Assistant) AI witness statement drafting",
    "technology_slug": "ai-report-writing",
    "vendor": "Anathem",
    "vendor_slug": "",
    "programme": "",
    "status": "paused",
    "stage": "in-use",
    "evidence": "academic",
    "evidence_is_claim": false,
    "since": "",
    "legal_basis": "",
    "oversight": "Independent evaluation by the Open University's Centre for Policing Research and Learning.",
    "notes": "An Open University evaluation dated 23 March 2025 examined a Hertfordshire Constabulary pilot of ADA (Anathem Digital Assistant), developed by Anathem, which takes the audio of Rapid Video Response contacts with domestic abuse victims, transcribes it and generates an MG11 witness statement, a DARA risk assessment and handover summaries; the report says ADA does not itself make decisions such as a final risk assessment. The report says the project was on 'pause' at the time of the study, December 2024 to March 2025, and gives no start date for the pilot. The authors found no consistent evidence that the first version of ADA saved time: drafts were compiled quickly, but call durations were slightly longer and report generation and checking took significantly longer, and they describe the productivity evidence as 'mixed (inconclusive)'. They found few spelling or grammar errors but lower readability in the AI statements, and write that ADA 'does not currently produce work that matches the average standard produced by the police officers we compared ADA with'. They report that hallucinations were present at the start of the work and were reduced 'to a great extent', but that a risk of hallucinations remains. Their overall conclusion is that version 1 did not provide proof of concept; they say version 2 was ready for testing and that there is 'a case to continue with this work'. Police Oracle reported the pilot on 20 February 2025. No source has been found saying whether the force resumed or ended use after March 2025; the status is recorded as paused on the strength of the report's statement, as at March 2025.",
    "source": "Open University Centre for Policing Research and Learning, evaluation report, 23 March 2025",
    "source_url": "https://university.open.ac.uk/centres/policing/sites/www.open.ac.uk.centres.policing/files/files/Outputs/Herts%20AI%20Report%202025.pdf",
    "last_checked": "2026-11-04",
    "record_url": "https://www.policeainews.com/tracker/record/fc726910-7daf-4fce-9d19-9b6ac7ea12ae"
  },
  "licence": "https://www.policeainews.com/data#licence"
}