AI GRCAPractice Lab
Step 5 of 5Evidence
05 · Practitioner output · Education

Set the responsible launch gate

Demo mode — sign in to save your work

Your recommendation should show whether learner fairness, accessibility and oversight evidence supports the proposed use.

PRACTICE MODELoading

Restoring your saved mode and lab work…

EDUCATION WORKING CONTEXT

Write the education decision record

The system supports learner engagement, assessment or safeguarding. Age, accessibility, educational fairness and appropriate staff judgement matter.

Evidence bar
  • Decision memo
  • Oversight sign-off
  • Learner outcome monitoring plan
REVIEW CHECKPOINTS

Responsible AI Recommendation

Responsible AI Recommendation

Complete the fields against the education operating context. Your work saves automatically in this browser.

Write at least 20 characters of your own answer to unlock the review guide.

Write at least 20 characters of your own answer to unlock the review guide.

Write at least 20 characters of your own answer to unlock the review guide.

Write at least 20 characters of your own answer to unlock the review guide.

Loading saved answers…

ASSURANCE WORKPAPER

Test whether the controls actually operate

Move beyond policy design: define a population, select a defensible sample, test evidence, document exceptions and write the conclusion an audit committee would need.

01Define

Objective, population and period.

02Sample

Risk-based and representative items.

03Test

Inspect, observe and reperform.

04Conclude

Rate findings and communicate.

STRUCTURED ARTEFACT

Risk → control → evidence traceability

Build an auditable decision trail. A risk without a control—or a control without evidence—remains visible as a gap.

No traceability rows yet. Add the first priority risk and connect it to a control, test and evidence.

SCENARIO INJECT

The situation has changed

Choose an event and update the governance response as if the system were already operating.

NEW EVENT

Unannounced model change

The supplier upgrades the underlying model and cannot confirm whether prior validation remains representative.

The business wants to keep the original approval because users have not complained.

Write at least 40 characters to unlock the response criteria.

PRACTITIONER MATURITYLevel 1 · Initial

Work has started, but material evidence and ownership are missing.

Why this level?
  • Workflow is 0% complete.
  • Worksheet answer coverage is 0%.
  • No complete risk-control-evidence row is ready for review.
  • Rubric ratings are not yet consistently strong or competent.
READINESS GATENot ready
Workflow
0%
Answers
0%
Traceability
0 rows
Why this gate?
  • Workflow completion is 0%.
  • Answer coverage is 0%.
  • Add at least one traceability row marked ready for review.
TRAINER FEEDBACK

What to strengthen next

  • Name the affected people and explain the pathway from system behaviour to impact.
  • Assign an accountable human role; responsibility cannot sit with the AI system.
  • Define the evidence and acceptance criteria needed to support the conclusion.
  • Add post-deployment monitoring, escalation thresholds and a reassessment trigger.
SELF-ASSESS BEFORE EXPORT

Practitioner quality rubric

Rate the work honestly. “Needs revision” is a useful result when it identifies what to strengthen before review.

Context and scope

Strong: States the system, decision, actors, people affected, boundaries and material assumptions.

Improve: Avoid generic statements or conclusions that depend on unstated facts.

Risk reasoning

Strong: Links a credible cause and event to a specific impact, then assesses likelihood, severity and uncertainty.

Improve: Do not list harms without explaining how they could arise in this scenario.

Proportionate controls

Strong: Selects preventive, detective and corrective controls that address the identified risks and assigns owners.

Improve: Avoid control lists with no connection to a risk, trigger or responsible role.

Evidence and decision

Strong: Defines testable evidence, acceptance criteria, residual risk and a clear, conditional recommendation.

Improve: A confident conclusion is not defensible when evidence, thresholds or escalation routes are missing.

EVIDENCE-QUALITY GATES
Practitioner evidence pack

PrinciplesAI

BrightPath Learning · Education overlay · Responsible AI principles review

Prepared by
Learner name not added
Course coverage
Week 2
Training simulation only · Fictional scenario and data

Executive assignment

BrightPath Learning plans an AI assistant that combines attendance, platform activity, assessment history and message sentiment to suggest which adult learners may need additional support. Tutors review the risk band and decide whether to contact a learner.

In simple terms: The education provider wants AI to spot learners who may be struggling so tutors can offer help. Tutors decide what to do, but the score could misunderstand a learner or treat some groups unfairly.

Assignment: Assess possible harms and decide when the proposed intervention should be paused or blocked.

Sector overlay: Education

Decision pressure: The steering committee meets in 10 working days. The launch slot will be lost if the decision is deferred beyond this meeting.

Ownership gap: Product, Risk and Operations each believe another function owns final residual-risk acceptance. The governance charter is silent.

Known facts

  • The assistant uses attendance, activity, assessment and message data
  • A sentiment model contributes to a learner risk band
  • The score does not directly determine grades or disciplinary action
  • Tutors decide whether and how to contact a learner
  • The reported 84% accuracy has no learner-group breakdown

Stakeholder Map

Stakeholder map
Not completed
Affected people and vulnerabilities
Not completed
Expected benefits
Not completed
Tensions and trade-offs
Not completed

AI Harm Assessment

Primary harm level
Not completed
Priority harm scenario
Not completed
Probability
Not completed
Severity
Not completed
Accuracy versus fairness
Not completed
OECD / trustworthy AI mapping
Not completed

Mitigation Plan

Primary treatment
Not completed
Controls
Not completed
Meaningful human oversight
Not completed
Evidence of effectiveness
Not completed

Responsible AI Recommendation

Launch recommendation
Not completed
Reasoning
Not completed
Launch blockers and red lines
Not completed
Accountable decision owner
Not completed
Post-launch monitoring
Not completed

Assurance testing workpaper

Assurance objective
Not completed
Population and period
Not completed
Sampling approach
Not completed
Test procedure
Not completed
Exceptions and root cause
Not completed
Finding and severity
Not completed
Management action
Not completed
Board / audit summary
Not completed

Quality review and sign-off

Reviewer: Responsible AI committee

Decision challenge: Does the assessment identify who may be harmed and set a credible launch-blocking threshold?

Context and scope
Not rated
Risk reasoning
Not rated
Proportionate controls
Not rated
Evidence and decision
Not rated

Reviewer sign-off: ____________________   Date: __________   Version: 1.0

Prepared as a learning artefact. Validate legal and regulatory conclusions against current authoritative sources before real-world use.
TRAINER REVIEW

Sign in to submit this evidence

Demo work stays in this browser. A registered account can submit a fixed version to a cohort trainer.

Sign in or create account
Back Finish lab