From the Public Safety edition of September 15, 2026
Criminal-justice AI framework ties adoption to validation, procurement gates and annual reassessment
Council on Criminal Justice Task Force on Artificial Intelligence · Criminal justice (law enforcement, courts, corrections, community) · United States (national guidance)
- Publisher
- Council on Criminal Justice
- Original publication
- March 2026
- Source retrieved
- 2026-09-15
What happened
The March 2026 CCJ user decision framework translates criminal-justice AI principles into a five-phase workflow: problem definition and readiness; prohibited-use screening and risk/opportunity classification; procurement protections; Level 1/Level 2 implementation safeguards (including pilots and override documentation); and ongoing monitoring with at least annual reassessment for substantial-risk systems. It explicitly requires independent validation rather than vendor claims alone and addresses staff use of general-purpose AI tools. It does not present measured outcome results from a specific deployment.
Why it matters
Directly usable by U.S. state and local public-safety agencies evaluating police, court, or corrections AI procurements. Provides shared language for sales, engineering, and delivery to structure go/no-go gates without inventing ROI.
Evidence and measured results
Full framework text retrieved from the CCJ publication page and cross-checked against the March 2026 PDF introduction/principles and phase structure.
Limitations and uncertainty
Guidance only; Microsoft listed among funders; local capacity may not support full Level 2 controls; promised 2026 case studies not yet available in this inspection.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-15; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Agencies lack a shared, justice-specific path to decide which AI tools to buy, pause, or refuse.
- Stakeholders
- agency executives
- procurement/legal
- IT/project managers
- community advisors
- oversight bodies
- Discovery questions
- Has the problem and non-AI alternatives been documented before vendor selection?
- Who approves substantial-risk classification memos?
- Can contracts require independent validation, audit logs, and termination rights?
- Value hypothesis
- Using the CCJ phases reduces avoidable procurement and rights risk by forcing validation and go/no-go gates before scale.
- Potential engagement
- Facilitated readiness and classification workshop producing a draft memo and procurement checklist tailored to one use case.
- Unsupported claims
- That the framework itself improves rearrest or clearance rates
- Guaranteed fairness after checklist completion
- Fitness of any specific vendor product
Pre-sales engineering
Role takeaway
- Fit
- Purpose-built criminal-justice AI under formal procurement, plus policy design for general-purpose staff AI tools.
- Architecture and integration
- Design recommendation UIs with override capture, demographic performance dashboards, and exportable audit trails into existing justice systems.
- Prerequisites
- Data governance basics
- Legal review capacity
- Defined success metrics and baselines
- Constraints
- Small-agency capacity limits for Level 2 controls
- Black-box systems may force higher risk class
- Framework is advisory, not regulation
- Security
- Encode confidentiality, access control, and data-rights terms before integration; prevent silent repurposing of collected data.
- Proof of value
- Independent validation plan on representative local data
- Pilot compares AI to current practice on pre-registered metrics
- Override logs reviewable for legal/appeals use
Delivery
Role takeaway
- Work
- Run the five-phase workflow for a named use case, including pilot evaluation and monitoring setup.
- Dependencies
- procurement
- counsel
- end-user supervisors
- community engagement lead for substantial risk
- Ownership
- Agency head or designee for classification approval and deployment authorization.
- Skills and adoption
- Mandatory user training on limitations and automation bias, with refreshers on model change.
- Governance checkpoints
- Phase exit approvals
- Contract execution before implementation
- Annual reassessment for substantial-risk systems
- Acceptance criteria
- Classification memo filed with risk/opportunity rationale
- Pilot go/no-go documented against success metrics
- Monitoring produces error and demographic performance reports
- Risks
- Shadow use of general-purpose AI on case data
- Vendor resistance to audit/termination terms
- Checklist completion mistaken for outcome proof
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Build for explainability/override paths, audit trails, and monitoring hooks before integration with RMS/CAD/court CMS/supervision systems; treat general-purpose chat tools as a separate governance surface from purpose-built procured systems.
Governance
Who approves, reviews and stays accountable for outcomes?
Require classification memos, multidisciplinary teams, community input for substantial-risk tools, and formal go/no-go after pilots. Annual reassessment should be budgeted, not optional.
Security and privacy
What data, permissions and controls need testing?
Procurement and implementation phases emphasise data rights, confidentiality, least collection, and access controls for sensitive justice data; contracts should encode auditability and termination rights.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Mandatory training and continuing education on system limits are core; community advisory participation for substantial-risk tools requires time and possibly compensation.
Procurement
What should contracts, pricing and exit terms secure?
Phase 3 treats contracts as the safety net: performance standards, fairness, auditability, data rights, and termination should be negotiated before acquisition.
Operating model
Which teams own the service once it runs?
Multidisciplinary assessment and annual reassessment become standing operating functions, not one-off project tasks; informal chatbot use needs a parallel policy.
Publication history
- 2026-09-15Public Safety · Issue 094 resources
Stable resource ID: ccj-ai-user-decision-framework-202603