{"resourceId":"ny-courts-ai-annual-report-2025-pilots","versions":[{"version":"external-776c65dbe01ae06d562ff96b74df85f55386aec0f4e34ec9dc8c12ad43a4944d","resource":{"id":"ny-courts-ai-annual-report-2025-pilots","title":"New York court report distinguishes pilot feedback, task suitability and security approval","organization":"New York State Unified Court System Advisory Committee on AI and the Courts","sector":"Courts","geography":"New York, United States","publishedAt":"Report dated December 2025; exact public release date unknown","publicationDate":null,"eventDate":null,"sourceName":"New York State Unified Court System","sourceLabel":"Official committee report with operator pilot accounts and policy appendices","sourceUrl":"https://webfiles.nycourts.gov/public/2026-01/ai-annual-report-2025.pdf","evidenceClass":"government-evaluation","outcomeClass":"emerging","topics":["knowledge-work","developers-agents","infrastructure","data-security","governance-procurement","accessibility-workforce","operating-model"],"finding":"The committee reports positive pilot feedback while distinguishing technical approval from suitability for particular court tasks.","sledRelevance":"Historical state-court deployment evidence fills a gap between general judicial interviews and actual institutional approval workflows.","evidence":"The report describes a two-month Copilot Chat pilot with more than 200 participants, nearing completion at report time. Feedback is qualitative; no controlled baseline, error denominator or measured net time saving is supplied.","architectureImplications":"Interpretation: Separate tenant configuration approval from validation of each permitted workflow.","governanceImplications":"Interpretation: Make a task-specific approval matrix and identify accountable judicial or administrative reviewers.","securityPrivacyImplications":"Interpretation: Test actual access, retention and connector behavior; avoid treating enterprise branding as a confidentiality guarantee.","caveats":"Selected pilot, policy and hosting sections inspected, not all 154 pages. Present rollout and vendor security status are unverified. Broad privacy assurances are operator assertions, not independent security findings.","streamIds":["public-safety"],"roles":{"sales":"Interpretation: Court administrators, judicial leaders, librarians, security officials and procurement teams face different reasons to consider assistance. Ask which task is approved, what employees currently do with sensitive records and who can judge output adequacy. A bounded engagement could evaluate one internal administrative workflow using authorized historical material. The value hypothesis is reducing avoidable handling while preserving review quality. The pilot account supports discovery questions, not a promise of faster adjudication or validated savings. Do not sell a general chat license as authorization for legal analysis. Applicability depends on each court's current policies, records environment and capacity to supervise the proposed task.","engineering":"Interpretation: Begin with the exact tenant, identity groups, connectors and data classes, then map how outputs return to the court system. Prerequisites include approved test documents and a court-owned reference set. Test cross-case access, misleading summaries, missing qualifications and unsupported citations where relevant. Compare total task completion and independently scored errors with the unassisted process. A model's hosting location does not establish that its output is reliable. Keep developer copilots subject to code review and keep workflow agents from filing or changing case records without authorized review. Recheck service versions and contract terms rather than assuming the historical product list remains current.","delivery":"Interpretation: A court administration sponsor should coordinate the technology division, judges, clerks, information security and training staff. Document the permitted workflow, stage a reversible pilot and reserve reviewer time before onboarding users. Dependencies include current approval, representative records and a staffed escalation route. Proposed acceptance: every test output remains traceable to authorized inputs, cross-case access tests fail safely, and critical errors receive documented correction before use. Set workload and accuracy targets locally in advance. Provide accessible training and human alternatives. Risks include confusing technical availability with professional suitability, silent changes in vendor behavior and scaling before downstream review costs are understood."},"retrievedAt":"2026-09-14T03:00:57Z","enrichedAt":"2026-09-14T03:03:45Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: Include clerks and assistive-technology users in evaluation and preserve assisted human service.","procurementImplications":"Interpretation: Require current evidence for the exact service configuration and include review labor in cost estimates.","operatingModelImplications":"Interpretation: Maintain distinct technical, legal-practice and operational acceptance responsibilities.","updateExplanation":"New-to-archive historical report; adds a state judiciary's pilot and approval experience. Report-time plans are not represented as September 2026 facts.","sourceVerification":{"openedUrl":"https://webfiles.nycourts.gov/public/2026-01/ai-annual-report-2025.pdf","referenceExcerpt":"any content generated by AI should be independently verified for accuracy.","promptVersion":"sled-research-v3.2","model":null,"basis":"agent-reported inspection"}}}]}