# AI / Algorithmic Impact Assessment (multi-framework)

> Create a cross-mapped AI impact assessment packet covering Canada AIA, EU AI Act Art. 27 FRIA, NIST AI RMF, NYC LL 144, UK ATRS, ICO AI auditing, and residual-risk register outputs.



Tags: AI, GRC, NIST AI RMF, EU AI Act, Algorithmic Impact Assessment, Public Sector, Bias Audit


## Example Prompts

- Create an AI impact assessment packet for our public benefits eligibility model
- Draft a NIST AI RMF profile, EU FRIA, Canada AIA score, and UK ATRS record for this AI system
- Prepare a NYC Local Law 144 bias audit summary and residual-risk register for our recruiting AEDT

URL: https://rakenne.app/skills/ai-algorithmic-impact-assessment/index.md

Try this skill: https://rakenne.app/a/?skill=ai-algorithmic-impact-assessment


## Overview

Create a cross-mapped AI governance packet from one system intake. The skill helps public-sector teams, regulated enterprises, and AI governance owners prepare Canada-style AIA scoring, EU AI Act Article 27 FRIA narrative, NIST AI RMF profile, NYC Local Law 144 bias-audit summary, UK ATRS transparency record, residual-risk register, and framework crosswalk.

## What It Covers

- Applicability routing across Canada AIA, EU FRIA, NIST AI RMF / AI 600-1, UK ATRS, NYC LL 144, ICO AI auditing, and ISO/IEC 42001 evidence.
- Evidence mapping for system purpose, data provenance, affected persons, fairness testing, human oversight, notices, recourse, monitoring, and reassessment triggers.
- Clear separation between publication-ready bias-audit summaries and methodology/data-request outputs when quantitative evidence is missing.

## Example Conversation

> **You:** We are a provincial benefits agency piloting an AI triage model for housing applications. It ranks cases for staff review and could affect low-income families and people with disabilities. Can you create the impact assessment packet for procurement and privacy review?
>
> **Agent:** I’ll build a multi-framework packet and mark missing evidence rather than inventing scores. I’ll start with the system role, affected groups, data sources, human review process, and available performance or fairness evidence.
>
> **You:** We have purpose, data sources, and staff-review workflow, but no subgroup fairness testing yet.
>
> **Agent:** I’ll produce a draft with a provisional Canada-style AIA rationale, EU FRIA assumptions, NIST AI RMF profile, UK ATRS-style transparency record, residual-risk register, and a data request for fairness metrics and monitoring evidence.

## Sample Output Excerpt

Excerpt from a generated assessment packet:

---

### Executive Summary and Applicability Matrix

The Housing Triage Assistant ranks incoming benefit applications for caseworker review. The system does not make final eligibility decisions, but it may affect service speed and prioritization for low-income applicants, persons with disabilities, and families with children.

| Framework              | Applies?            | Rationale                                                                                           | Deliverable status                              |
| ---------------------- | ------------------- | --------------------------------------------------------------------------------------------------- | ----------------------------------------------- |
| Canada AIA             | Yes / provisional   | Public-sector automated decision support affecting benefit administration.                          | Draft score with evidence gaps                  |
| EU AI Act Art. 27 FRIA | Possible future use | If deployed by an EU public body or essential-service provider, deployer-side FRIA may be required. | Assumptions documented                          |
| NIST AI RMF            | Yes                 | Enterprise AI governance baseline.                                                                  | Govern / Map / Measure / Manage profile drafted |

### Residual Risk Register

| Risk                                                            | Affected group                                 | Mitigation                                                               | Residual risk                               | Owner         | Reassessment trigger                                    |
| --------------------------------------------------------------- | ---------------------------------------------- | ------------------------------------------------------------------------ | ------------------------------------------- | ------------- | ------------------------------------------------------- |
| Delayed review for applicants underrepresented in training data | Low-income families, persons with disabilities | Human caseworker review, queue monitoring, subgroup testing data request | Medium until fairness evidence is available | Program owner | New model release or monthly disparity threshold breach |

<!-- /excerpt -->

## Extension Tools

**`multi_framework_aia_check`** validates a completed packet for required deliverables and professional guardrails:

- Canada-style AIA, EU Art. 27 FRIA, NIST AI RMF, NYC LL 144, UK ATRS, residual-risk register, and crosswalk coverage.
- Evidence-gap handling so unsupported scores or fairness metrics are not presented as findings.
- Human oversight, complaint/appeal/recourse, affected groups, monitoring, and reassessment trigger coverage.
- NYC LL 144 distinction between publication-ready summaries and methodology/data-request outputs.

## Getting Started

Bring the AI system description, deployment context, affected populations, jurisdictions, available data/performance/fairness evidence, and any existing risk register or technical documentation. If quantitative bias-audit data is not available yet, the skill will produce a defensible evidence request instead of filling in unsupported numbers.



---

Back to [Skill Library](https://rakenne.app/skills/index.md)