Section 9: Certification Tiers

Three levels of trust, matched to three levels of risk.

9.1 Overview

AI tools used in courts vary by function and risk. A scheduling assistant does not present the same risk profile as a tool that generates sentencing recommendations. The BenchMark Standard recognizes this reality through a three-tier certification model.

Each tier defines:

Which domains must be passed (and at what threshold).
What score thresholds apply.
What judicial functions the tool is certified for.
What ongoing obligations attach to the certification.

Certifications under the BenchMark Standard are issued by the Judicial AI Standards Institute, the certifying body described in Section 1.7. They are not issued by individual courts and are not issued by the adopting authority. A certification represents the result of an independent evaluation conducted under the published methodology by a qualified evaluator. Once issued, the certification appears on the public registry maintained by the Institute at judicialaistandards.org and on the list published by the adopting authority.

Threshold Calibration Note: The score thresholds (75 for Verified and Certified, 90 for Certified-Sensitive) are calibrated to the scoring rubric in Section 2.3: a score of 75 represents "Meets" (satisfies all minimum requirements), while 90 represents "Exceeds" (surpasses minimums). These thresholds reflect professional judgment informed by the scoring guide worked examples and the framework's design principles. As more tools are evaluated under this framework, thresholds will be reviewed and adjusted based on empirical data in future versions.

What Certification Decides and What It Does Not Decide: Certification confirms that a tool meets the substantive standards of the six domains and is technically capable of being used consistently with applicable Tennessee law and rule. Certification does not decide whether a court must disclose AI use in a particular matter, what notice a party is entitled to receive, what record must be preserved for appellate review, or how a party-challenge motion must be adjudicated. Those determinations are governed by Tennessee law, court rule, local rule, standing order, and the supervisory authority of the adopting authority and the presiding judge. Section 5.5 (Disclosure Support, Notice Capacity, and Record Preservation) describes the tool capabilities that certification preserves so those determinations can be made on a complete record. Section 1.2 (Real-Time Courtroom Use) describes the boundary the certification holds at the courtroom door: tools must be capable of being used consistently with the Tennessee Rules of Evidence, Criminal Procedure, Civil Procedure, Juvenile Practice and Procedure, Supreme Court Rules, and Code of Judicial Conduct, but the framework does not regulate courtroom behavior.

9.2 Tier 1: BenchMark Verified

What It Means

The tool meets baseline safety standards for administrative and clerical use in court settings. It handles data responsibly, provides transparent outputs, and keeps humans in control.

Requirements

Domain	Requirement
Domain 1: Accuracy	Score ≥ 75
Domain 2: Bias	Score ≥ 55
Domain 3: Constitutional	Score ≥ 55
Domain 4: Security	Score ≥ 75, no critical failures
Domain 5: Transparency	Score ≥ 75
Domain 6: Human Override	Score ≥ 75, no critical failures

Approved Uses

Calendar and scheduling management.
Document formatting and template generation.
Case management data entry assistance.
Docket management and tracking.
Administrative correspondence drafting.
Court statistics and reporting.

Prohibited Uses

Legal research or analysis.
Drafting orders, judgments, or rulings.
Risk assessment of any kind.
Recommendation or decision support.
Any function a judge relies on for substantive decisions.

Recertification

Annual evaluation.
No mid-cycle recertification required for model updates (administrative tools).

Mandatory Disclosure

Verified-tier tools must achieve at least a 55 in Domain 2 (Bias) and Domain 3 (Constitutional Compliance). This is a floor, not a passing threshold; full passage at 75 is required only for Certified and Certified-Sensitive tiers. Verified tools that score between 55 and 74 in Domains 2 or 3 must include in all certification displays: "Bias and constitutional compliance evaluated; floor met; full report available." Courts adopting Verified tools should review the Domain 2 and 3 scores in the evaluation report before deployment, particularly for tools that affect notice, access, scheduling, language services, or public-facing court information, where bias and due process considerations can arise even in administrative contexts.

Certification Mark

Tools receiving Tier 1 certification may display:

BenchMark Verified™.
Certified for administrative court use.
Issued by the Judicial AI Standards Institute.
Verify at judicialaistandards.org/registry.
Valid through [date].

9.3 Tier 2: BenchMark Certified

What It Means

The tool meets the full BenchMark Standard across all six domains. It is safe for integration into judicial workflow (legal research, drafting assistance, analytics, and decision support) with the standing requirement that all outputs are reviewed by a judge or licensed attorney.

Requirements

Domain	Requirement
Domain 1: Accuracy	Score ≥ 75
Domain 2: Bias	Score ≥ 75
Domain 3: Constitutional	Score ≥ 75
Domain 4: Security	Score ≥ 75, no critical failures
Domain 5: Transparency	Score ≥ 75
Domain 6: Human Override	Score ≥ 75, no critical failures

Approved Uses

All Tier 1 uses, plus:

Legal research and citation checking.
Drafting assistance for orders, memoranda, and opinions.
Case law analysis and comparison.
Workflow automation (with human approval gates).
Statistical analysis and pattern identification.
General decision support (advisory, never determinative).

Mandatory Conditions

All AI-generated legal analysis must be reviewed by a judge or licensed attorney before use.
The tool must clearly label all outputs as AI-generated.
Courts must maintain audit logs of all AI interactions.
Users must be trained on the tool's limitations.

Recertification

Annual full evaluation.
Recertification required within 90 days of any major model update.
Vendor must notify all certified courts of model changes within 30 days.

Certification Mark

BenchMark Certified™.
Certified for judicial workflow integration.
All outputs require human review.
Issued by the Judicial AI Standards Institute.
Verify at judicialaistandards.org/registry.
Valid through [date].

9.4 Tier 3: BenchMark Certified-Sensitive

What It Means

The tool meets the highest BenchMark Standard and is safe for use in proceedings involving juveniles, sealed records, mental health, substance abuse, and other sensitive categories where privacy and constitutional protections are heightened.

Requirements

Domain	Requirement
Domain 1: Accuracy	Score ≥ 90
Domain 2: Bias	Score ≥ 90
Domain 3: Constitutional	Score ≥ 90
Domain 4: Security	Score ≥ 90, no critical failures
Domain 5: Transparency	Score ≥ 90
Domain 6: Human Override	Score ≥ 90, no critical failures

Additional Testing

Beyond the standard six-domain evaluation, Tier 3 requires:

Juvenile-Specific Testing:

25 additional juvenile case scenarios (delinquency, dependency, status offenses).
Confidentiality stress testing (attempts to extract juvenile identities).
Age-appropriate language and recommendation testing.
DCS interaction scenarios.

Termination and Adoption Testing:

10 termination of parental rights scenarios across juvenile, chancery, and circuit court postures.
10 adoption scenarios including consent matters and contested adoptions.
TPR statutory grounds analysis under T.C.A. § 36-1-113, including all enumerated grounds.
Best-interest factor analysis specific to TPR and adoption proceedings. In TPR and adoption matters, the best-interest inquiry is governed by the statutory framework for termination of parental rights and adoption. See Tenn. Code Ann. § 36-1-113 (TPR grounds and best-interest framework); Tenn. Code Ann. tit. 36, ch. 1 (adoption). It is a separate inquiry from the custody best-interest factors that apply in ordinary parenting-plan or custody disputes under Tenn. Code Ann. § 36-6-106(a).
Confidentiality stress testing for adoption records under T.C.A. § 36-1-126.

Sealed Record Testing:

15 sealed record scenarios across criminal, juvenile, and civil contexts.
Cross-case inference testing (can sealed data be deduced from non-sealed outputs?).
Long-term data leakage testing (does sealed data surface days or weeks later?).

Mental Health & Substance Abuse:

10 involuntary commitment scenarios.
10 substance abuse treatment court scenarios.
Health Insurance Portability and Accountability Act (HIPAA) and 42 C.F.R. Part 2 compliance verification.

Approved Uses

All Tier 1 and Tier 2 uses, plus:

Juvenile court case management and analysis.
Sealed record proceedings.
Mental health court workflows.
Drug/recovery court workflows.
DCS and child welfare case support.
Termination of parental rights proceedings in juvenile court, chancery court, or circuit court (T.C.A. § 36-1-113, concurrent jurisdiction).
Adoption proceedings in chancery court, including consent matters, contested adoptions, and adoptions following TPR.
Probate matters involving minor heirs or contested capacity determinations.
Guardianship and conservatorship proceedings under T.C.A. Title 34.
Any proceeding with heightened confidentiality requirements.

Mandatory Conditions

All Tier 2 conditions, plus:

Enhanced audit logging with daily review.
Quarterly security reviews.
Annual bias audit with published results.
Designated privacy officer oversight.
Incident response plan on file with court administrator.

Recertification

Full evaluation every 12 months.
Quarterly monitoring reviews (abbreviated evaluation of Domains 2, 4, and 5).
Immediate recertification upon any model change, data source change, or security incident.
Certification suspended pending recertification upon any reported data breach.

Certification Mark

BenchMark Certified-Sensitive™.
Certified for juvenile, sealed, and sensitive proceedings.
Enhanced monitoring and quarterly review.
Issued by the Judicial AI Standards Institute.
Verify at judicialaistandards.org/registry.
Valid through [date].

9.5 Certification Decision Matrix

Scenario	Tier Required
Court clerk uses AI to manage hearing calendar	Verified
Judge uses AI for legal research on a motion to dismiss	Certified
Staff uses AI to draft standard form orders	Certified
AI assists with juvenile delinquency risk assessment	Certified-Sensitive
AI manages case files in drug court	Certified-Sensitive
AI helps analyze sentencing disparity data	Certified
AI drafts juvenile transfer hearing memorandum	Certified-Sensitive
Court uses AI for public-facing FAQ chatbot	Verified
AI assists judge in evaluating DCS permanency plans	Certified-Sensitive
AI assists in TPR proceeding (any court)	Certified-Sensitive
AI used in adoption case file review	Certified-Sensitive
AI assists chancery court divorce drafting	Certified
AI assists chancery court probate of contested estate	Certified
AI used in guardianship of minor proceedings	Certified-Sensitive

9.6 Multiple Certifications

A single tool may receive different tier certifications for different functions. For example:

A case management system's scheduling module might receive Verified certification.
The same system's legal research module might receive Certified certification.
The same system's juvenile case analysis module might receive or fail to receive Certified-Sensitive certification.

Each function is evaluated independently. The tool's marketing may only reference the tier achieved for each specific function.

9.7 Certification Suspension and Revocation

Certifications can be suspended or revoked when:

Event	Action
Unpatched critical security vulnerability	Immediate suspension
Confirmed data breach involving court data	Immediate suspension
Failed recertification	Revocation after 90-day remediation window
Vendor non-cooperation with monitoring	Suspension after 30-day notice
Material misrepresentation in application	Immediate revocation
Court-reported critical failure in production	Suspension pending investigation

Courts are notified of suspension or revocation within 24 hours via the BenchMark registry (V2).

9.8 The Trust Ladder

The three tiers create a natural progression:

New vendors start at Verified. Demonstrate baseline safety for low-risk functions.
Mature vendors achieve Certified. Prove full-spectrum compliance for judicial workflow.
Vendors that meet the highest tier earn Certified-Sensitive. Prove they can handle the most protected proceedings.

This ladder incentivizes continuous improvement. Vendors have a clear path to higher certification (and higher-value contracts with courts that require it).

Note on Companion Documents

This white paper presents the evaluation framework and scoring methodology. Companion documents, in development for accompaniment of the final submission to the Tennessee Administrative Office of the Courts, are:

Judicial AI Readiness Assessment. A self-assessment checklist for courts considering AI adoption, covering technical readiness, governance, and staff training.
Vendor Self-Assessment Checklist. A structured pre-submission instrument for vendors preparing for formal evaluation.
Court Implementation Guide. A plain-language operational reference for courts adopting BenchMark-certified tools, scoped to procurement and use. The Court Implementation Guide does not enable a court to evaluate a vendor submission; that role belongs to the certifying body. The Guide explains how to read a certification report, what the tier designations mean for procurement decisions, and what ongoing court obligations attach to deploying a certified tool.
Evaluator Scoring Guide. Sample scoring scenarios and detailed rubrics for evaluators conducting formal certifications.
Sample Evaluation Report template. The standard format for evaluation reports issued by the certifying body.