Copied to clipboard!
Module 5 For Government intermediate 35 min

Assess Before You Deploy: Evaluation & the PII/CJIS Guardrail

Rule 3 + the Evaluation Imperative

What you'll be able to do

  • Score your entity against the Five Rules and match the right Ascend 2030 pilot
  • Run a pre-deployment impact assessment before any pilot launches
  • Apply the PII and CJIS guardrail to every step that touches constituent data

The hard guardrail

NEVER enter into any AI tool: resident or applicant names, addresses, or unit numbers; SSNs, DOB, or government-issued ID numbers; case numbers, file IDs, or system identifiers; income, employment, or benefits tied to an individual; details of open grievances, hearings, or legal proceedings involving named individuals; system credentials; unredacted reports containing identifying information; and any criminal-justice information (CJIS).

ALWAYS anonymize before describing a specific situation. ALWAYS verify AI output against current policy before use. ALWAYS treat AI output as a draft, not a final decision. Use the PII Redactor tool to scrub documents to "Resident A" or "Case 1" before pasting anything.

Five Rules Readiness & Pilot Selection

Before you pitch, fund, or launch an AI pilot, you need an honest read on whether your entity is ready, and which pilot fits. This exercise walks you through scoring your department against the Five Rules of Government AI (0 to 2 each, max 10), then matching your score and your department mix to the right Ascend 2030 pilot template. The output is a one-page readiness brief you can take to a governance team or council.

Recommended tool

claude

Scoring and pilot-matching is reasoning over your described situation against a fixed rubric. Claude is strong at structured assessment and will produce a clean, defensible one-page brief. Keep it in one thread so you can adjust scores and immediately see how the pilot recommendation shifts.

Bring to the exercise

  • Your entity type and the departments in scope
  • Any current AI tools already in use (including vendor tools with embedded AI)
  • Existing policies: AI use registry, impact-assessment process, transparency/disclosure, human-review requirements
  • Your highest-volume document or service pain points
  • Available partners (planning/dev, public safety and community orgs, or health systems)
Step 1

Inventory Before You Innovate (Rule 1)

You cannot govern what you cannot see. Score Rule 1 honestly.

What to substitute before pasting

  • [DESCRIBE ...] Be honest. "We have no AI inventory" is a valid and common answer that scores 0.
Score my entity on Rule 1 of the Five Rules of Government AI: "Inventory Before You Innovate."

Rubric: 0 = no signal of this in our operations; 1 = some elements present, not formalized; 2 = published policy, documented practice, or operational evidence.

Here is what we have today: [DESCRIBE any AI tool inventory, vendor inventory, departmental AI catalog, or mentions of "AI inventory" in council minutes/audits, or say "none that I know of"].

Give me a 0/1/2 score, the one-line evidence behind it, and the single most valuable next step to raise the score.
What good output looks like
  • A defensible 0/1/2 with one-line evidence.
  • A concrete next step (e.g., "have each department head catalog embedded-AI tools in 2 weeks").

Verify before using AI's output

Step 2

Score Rules 2 to 5

Score the remaining four rules with the same rubric.

What to substitute before pasting

  • [DESCRIBE or "none"] One honest sentence per rule. Most entities score low on 1 and 4.
Now score Rules 2 through 5 on the same 0/1/2 rubric. For each, here is what we have:

Rule 2 (Train the Frontline, Not Just IT): frontline staff AI training (311, inspectors, clerks, caseworkers), not just IT: [DESCRIBE or "none"].
Rule 3 (Assess Before You Deploy): pre-deployment AI impact assessment; NIST AI RMF alignment: [DESCRIBE or "none"].
Rule 4 (Default to Transparency): public AI use registry; plain-language resident explanations: [DESCRIBE or "none"].
Rule 5 (Keep Humans in the Loop): documented human-review requirement for consequential decisions: [DESCRIBE or "none"].

For each rule give a 0/1/2 with one-line evidence. Then total the score out of 10 and tell me the readiness band:
- 8 to 10 high: ready for an advanced/multi-pilot engagement
- 5 to 7 medium: ready for ONE pilot
- 2 to 4 low: needs foundation work (readiness assessment + Five Rules workshop) before any pilot
- 0 to 1 not ready: a council briefing / education engagement, not a pilot
What good output looks like
  • Five scores, a total out of 10, and the matching readiness band.
  • The band drives the recommendation. It does not oversell a Tier-2 pilot to a low-readiness entity.

Verify before using AI's output

Step 3

Match the Pilot to Your Department Mix

Map your readiness and departments to one of the three Ascend pilots (or a custom sweet-spot use case).

What to substitute before pasting

  • [DESCRIBE] Name real departments and their real document/service volume.
Based on my score and my departments, recommend ONE Ascend 2030 pilot template:

- Pilot 1 (Intelligent Document Processing): best if we are permit/records-heavy (planning & development, code enforcement, building inspection, court records).
- Pilot 2 (Community Safety Planning): best if leadership has a public-safety priority, designated neighborhoods, and community-org infrastructure.
- Pilot 3 (Healthcare Corridor Data Intelligence): best if we have major regional health systems / a public-health department / an insurer presence.
- Custom: if none fit, propose ONE use case in the Playbook "sweet spot": low risk plus at least moderate benefit.

My departments and pain points: [DESCRIBE]. My partners available: [DESCRIBE].

Tell me the pilot, a two-sentence rationale tied to my specifics, the department partners to involve, and which cognitive-delegation pattern (AI Zone / Human Zone / Collaboration Zone) is most load-bearing for this pilot.
What good output looks like
  • One named pilot with a rationale that cites YOUR department signals, not vibes.
  • Named department partners and the key AI/Human/Collaboration split.
  • For a low-readiness score, it recommends foundation work first rather than rushing a pilot.

Verify before using AI's output