Ethical Review¶
An ethical review evaluates the fairness, privacy, safety, transparency, and societal-impact posture of an AI system, experiment, dataset, or model before deployment or publication. It is the FCC equivalent of an Institutional Review Board (IRB) check, extended to cover AI-specific concerns (disparate impact, dual-use, explainability gaps). Produce this artifact during the Critique phase for any artifact that processes personal data, makes decisions affecting individuals, or could be repurposed for harmful applications.
Template¶
Section 1: Review Metadata¶
Instructions: Ethical reviews are never anonymous. Record type (Pre-deployment / Pre-publication / Periodic / Incident-triggered), reviewer identity, date, and Innovation / Ecosystem ID.
| Field | Value |
|---|---|
| Review ID | [FILL — e.g. ETH-2026-001] |
| Reviewer | [FILL] |
| Date | [FILL] |
| Review type | [Pre-deployment / Pre-publication / Periodic / Incident-triggered] |
| Innovation / Ecosystem ID | [FILL] |
Section 2: Artifact & Intended Use¶
Instructions: A well-scoped intended use and explicit out-of-scope uses are the anchor for every subsequent ethical judgement. Vague intended-use statements invalidate the review.
| Field | Value |
|---|---|
| Artifact name / version | [FILL] |
| Artifact type | [Model / Dataset / Experiment / System / Agent / Other] |
| Authors | [FILL] |
| Intended use | [FILL] |
| Intended users | [FILL] |
| Out-of-scope uses | [FILL] |
Section 3: Risk Classification¶
Instructions: Tick the applicable EU AI Act risk tier and complete the NIST AI RMF Govern / Map / Measure / Manage rows. Tier Unacceptable is a hard-stop: the artifact cannot proceed.
- EU AI Act tier:
[Unacceptable (hard-stop) / High / Limited / Minimal] - NIST AI RMF — Govern:
[FILL] - NIST AI RMF — Map:
[FILL] - NIST AI RMF — Measure:
[FILL] - NIST AI RMF — Manage:
[FILL]
Section 4: Fairness, Bias, and Privacy¶
Instructions: Tick each item only when the corresponding evidence exists. Sensitive data requires an affirmative DPIA reference.
- Training / evaluation demographics documented
- Known biases identified and mitigated
- Fairness metrics defined and measured
- Disparate-impact analysis conducted (if applicable)
- Personal-data inventory completed
- Data minimisation applied
- Consent / legal basis documented
- Anonymisation / pseudonymisation where required
- DPIA completed or explicitly waived
Section 5: Safety, Robustness, Transparency¶
Instructions: Safety covers failure modes, human oversight, and rollback. Transparency is linked to the Model Card (OPEN-SCI-004a) and — for agents — the Agent Transparency Card (OPEN-SCI-009).
- Failure modes identified and documented
- Adversarial robustness tested where applicable
- Fallback / kill-switch mechanisms available
- Human-in-the-loop oversight defined
- Model Card completed (link to OPEN-SCI-004a)
- LLM usage declared (models, tasks, limitations)
- Agent Transparency Card completed (link to OPEN-SCI-009)
Section 6: Societal Impact, Dual Use, and Decision¶
Instructions: Enumerate positive and negative impacts (with likelihood and severity) and declare dual-use safeguards. The four decision labels are mutually exclusive; Approved with conditions requires explicit enumerate conditions.
- Positive impacts:
[FILL] - Negative impacts (likelihood / severity / mitigation):
[FILL] - Dual-use considerations + safeguards:
[FILL] - Decision:
[Approved / Approved with conditions / Requires revision / Rejected] - Conditions (if applicable):
[FILL] - Next review date:
[FILL]
Adoption Checklist¶
- All required sections completed
- Artifact peer-reviewed by at least one R.I.S.C.E.A.R. peer
- Stored in the project's designated docs location
- Linked from README or equivalent index
- Versioned + date-stamped with a scheduled re-review
References¶
- PHOENIX v4.0.0 —
docs/resources/templates/open-science/ethics-review.md - IEEE 7000-2021 — Ethical Design of Autonomous Systems
- NIST AI RMF 1.0 (2023) — AI Risk Management Framework
- EU AI Act (Regulation 2024/1689) — Risk Classification
- Montreal Declaration for Responsible AI (2018)
FCC integration¶
This template is referenced from the Forensic Auditor persona
(src/fcc/data/personas/forensic_auditor.yaml) as part of the
Critique-phase evidence set. Ethical reviews are cross-linked by the
compliance auditor under src/fcc/compliance/auditor.py to their
EU AI Act and NIST AI RMF requirements in
src/fcc/data/compliance/eu_ai_act_requirements.yaml and
src/fcc/data/compliance/nist_ai_rmf_mapping.yaml. See also
src/fcc/data/governance/ethics_framework.yaml and
src/fcc/data/governance/ethics_assessment.yaml.