# ACIN AI Policy and CLEST Training Packet This cluster contains: - `GOV-001_ACIN_AI_Fact_Sheet.pdf`; - `GOV-002_Arkansas_Model_AI_Policy_TEMPLATE.pdf`; - `GOV-201_Arkansas_Act_848_of_2025_AI_Policy_Authority.pdf`; - `GOV-202_Arkansas_AI_Acceptable_Use_Policy_Final_2026-02-26.pdf`; - `TRN-001_ACIN_CLEST_AI_Training_Packet_FINAL_SOURCE_2026-07-06.pdf`; - `BRF-401_AI_in_Law_Enforcement_2026-05-23.pdf` and its Word-format counterpart `BRF-401N_AI_in_Law_Enforcement_2026-05-23.docx`. The records distinguish public explanation, model policy, statewide policy, and a prepared training-submission package. They should not be collapsed into one adopted ACIN policy or an externally approved course. ## ACIN fact sheet and essay The fact sheet says AI supports pattern detection, cross-jurisdiction case linking, lead generation, trend analysis, and relationship analysis, with human control over final decisions. It prohibits autonomous arrests, AI-only probable cause, social scoring, unauthorized surveillance, replacement of human judgment, and protected-status-only profiling (`GOV-001_ACIN_AI_Fact_Sheet.pdf`, p. 1). The PDF/DOCX essay versions similarly argue that AI results require review, verification, bias monitoring, privacy protection, and human authority (`BRF-401_AI_in_Law_Enforcement_2026-05-23.pdf`, p. 1; `BRF-401N_AI_in_Law_Enforcement_2026-05-23.docx`, full document). ## Act 848 and statewide policy The produced Act 848 copy requires covered public entities to adopt AI/automated-decision-tool policies, requires an authorized human employee or designee to make final decisions, and requires employee training on the policy and appropriate AI/ADT use in deciding an outcome (`GOV-201_Arkansas_Act_848_of_2025_AI_Policy_Authority.pdf`, pp. 4-5). The official enrolled Act confirms those statutory minimums; its public-on-request clause expressly addresses each state entity's policies. (primary public record, [Arkansas Act 848 of 2025](../../../../web%20archive/2026-08-10/arkleg.state.ar.us/arkansas-act-848-of-2025.md)) The produced Arkansas statewide AI Acceptable Use Policy is marked final and effective 2026-02-26. It requires human review, approved tools, auditability, training, and security controls; it prohibits reliance on AI alone for binding decisions and surveillance, profiling, or tracking without explicit statutory authority (`GOV-202_Arkansas_AI_Acceptable_Use_Policy_Final_2026-02-26.pdf`, pp. 1-4). These are requirements of the produced statewide policy; they should not all be attributed to Act 848 itself. This record documents the statewide policy ACIN cited. It does not itself prove that every ACIN vendor feature completed the state's review and approval process. ## Model policy `GOV-002` is expressly labeled "Template for Agency Adoption." It requires local legal, security, privacy, procurement, vendor-risk, accuracy, bias, retention, discovery, and command review before acquiring or materially changing AI. It also requires human review and limits vendor access, training, resale, and secondary use (pp. 1-2). Because it is a template, it is not evidence that every participating agency adopted it. ## CLEST packet `TRN-001` is a 55-page package prepared for the [[Arkansas Commission on Law Enforcement Standards and Training]]. It includes: - a 25-slide presentation; - a submission manual and lesson plan; - instructor and student material; - tests, answer keys, sign-in and evaluation forms; - instructor certification and completion-certificate templates; - copies/summaries of Act 848, state guidance, the ACIN fact sheet, and the model policy. The course repeatedly states that AI assists and humans decide; it addresses hallucinations, bias, deepfakes, CJIS, audit logs, and prohibited autonomous decisions (`TRN-001_ACIN_CLEST_AI_Training_Packet_FINAL_SOURCE_2026-07-06.pdf`, pp. 9-25, 45-53). The administrative pages retain blank CLEST-number and approval fields, and the certificate is explicitly a template (`TRN-001_ACIN_CLEST_AI_Training_Packet_FINAL_SOURCE_2026-07-06.pdf`, pp. 48-51). Although template text uses phrases such as "CLEST-approved training program," the package contains no completed approval, assigned CLEST number, external approval letter, or executed certificate. It establishes a prepared submission/course package, not external CLEST approval. ## Contract relationship The human-review rules are consistent with Peregrine's contract addendum, which calls AI/ML decision-support and leaves law-enforcement decisions to human review ([[Peregrine and Carahsoft Executed Contract]], `CON-002_Peregrine_Carahsoft_Executed_Contract_2026-03-17.pdf`, p. 22). The LeadsOnline package, however, gives concrete narrative, pattern, phone, movement, and case-linking functions whose risk assessment and approval record are not produced. ## Significance The materials demonstrate that ACIN anticipated AI governance, human oversight, and training. They also show a difference between stated safeguards and proof of institutional adoption or system approval: templates, prepared courses, and contract clauses are not completed audits, approvals, or usage reviews. ## Open questions / follow-ups - Did CLEST approve the course, assign a course number, or authorize delivery? - Which ACIN and participating-agency AI policies were actually adopted, and on what dates? - Did OST or another state review body approve Peregrine, LeadSuite, NightHawk, or CellHawk? - Produce inventories, risk assessments, bias/accuracy tests, decision logs, training completion records, and periodic reviews. ## Current-law authority crosswalk The 2026-08-10 legal audit uses these authorities only for the bounded propositions stated below: - (primary public record, [HB 1958 / Act 848 bill history](../../../../web%20archive/2026-08-10/arkleg.state.ar.us/hb-1958-act-848-bill-history.md)) — Act identity, approval, and legislative history.