Intermediate
AI Ethics Review Board
An AI ethics review board (also called a Responsible AI Committee, AI Ethics Council, or AI Governance Board) is a body that provides oversight of an organisation's AI systems and practices from an ethical and values perspective. When well-designed, it catches risks that technical teams are too close to see and provides legitimacy for high-stakes AI deployments. When poorly designed, it becomes a rubber-stamp that legitimises harmful AI while consuming resources and creating a false appearance of governance.
Charter and Mandate
Before establishing the board, the organisation must define what authority it actually has. An ethics review board without defined authority is a discussion group, not a governance body:
Charter elements
- Scope: What AI systems and decisions does the board review? Typically: high-risk AI systems (Annex III categories), new AI use cases in sensitive domains, AI incidents with significant ethical dimensions, and requests for ethics guidance from product teams
- Decision authority: Can the board block a deployment? Require mitigation? Or only advise? The answer must be written into the charter — advisory-only boards with no blocking authority are structurally limited
- Escalation path: If the board recommends against a deployment and the business unit disagrees, who resolves the conflict? Typically: CEO or Board of Directors for material disagreements
- Reporting line: Board should report to the CEO or Board — not to the CTO or product division, which creates a conflict of interest
- Sunset review: Charter should be reviewed annually — AI risk landscape and regulatory requirements evolve
Board Composition
| Role | Why included | Typical representation |
|---|---|---|
| AI/ML engineers | Technical feasibility of mitigations; understanding of model capabilities and limitations | 1-2 senior engineers; rotated to avoid capture |
| Legal / compliance | Regulatory obligations; liability assessment; contract implications | Legal counsel; compliance officer |
| Ethics / social science | Ethical analysis beyond legal compliance; values reasoning; societal impact framing | Internal ethics team lead or external academic |
| Domain expert | Contextual expertise in the deployment domain (healthcare, finance, HR) | Varies by use case; may be an ad hoc SME rather than standing member |
| External advisors | Independence from commercial interests; diverse perspectives; legitimacy with external stakeholders | Civil society, academia, independent technologists |
| Affected community representation | Direct experience of AI decision impact; perspectives internal teams lack; accountability to affected people | Community advocates; user representatives; rarely included but most important |
Review Process
The ethics review process should be structured enough to be consistent and defensible, but flexible enough to adapt to diverse AI systems:
- Intake and triage: Product team submits a review request with system description, intended use, deployment scope, and preliminary risk assessment. Board secretariat triages to determine review depth required.
- Risk assessment preparation: Product team completes an AI Impact Assessment (aligned to ISO 42001 Clause 8.4) — potential harms, affected groups, deployment context, proposed mitigations.
- Board review meeting: Product team presents; board members ask questions; ethics dimensions assessed against charter principles; external stakeholder input considered where appropriate.
- Decision and conditions: Board approves (with or without conditions), defers pending additional information, or recommends against deployment. Conditions must be specific and verifiable.
- Implementation verification: For conditional approvals, board verifies conditions were met before deployment — not a rubber stamp; spot audits of implemented mitigations.
- Post-deployment review: High-risk systems reviewed 6 months after deployment for intended use violations and unanticipated harms.
Common Failure Modes
Structural failures
- Capture: Board composition dominated by employees with commercial incentives — external voices are token rather than influential.
- Advisory-only with no blocking authority: Business units can ignore recommendations. The board becomes a documentation exercise, not a governance mechanism.
- Late-stage review: Ethics review happens after significant development investment — making it psychologically and economically difficult to reject anything. Review must happen early.
Process failures
- Ethics washing: Board approves everything with minor recommendations; no high-risk system has ever been blocked. Board existence is used externally as evidence of responsibility without substantive review.
- Scope limitation: Board only reviews new systems; existing systems with significant ethical issues never come up for review.
- No follow-through: Conditional approvals with conditions that are never verified.
Effective Ethics Review in Practice
- Microsoft: Sensitive Use Review process requires all uses of Azure AI services in high-risk categories (facial recognition, healthcare, criminal justice) to receive approval before deployment. Involves both technical and ethics review; can require additional mitigations or deny permission to proceed.
- Google: PAIR (People + AI Research) and the internal Responsible Innovation team provide ethics review and design guidance. The dissolution of the external AI Ethics Board in 2019 (due to composition controversy) is a cautionary example of external board governance challenges.
- Regulatory expectation: The EU AI Act requires "appropriate governance structures" for high-risk AI systems. While not mandating a specific board structure, ethics review documentation is expected as part of technical documentation and conformity assessment.
Checklist: Do You Understand This?
- What authority must an ethics review board have to be genuinely effective — what is the difference between advisory and decision authority?
- Why should the board report to the CEO or Board rather than to the CTO or product organisation?
- What is ethics washing, and what structural feature of a review board indicates it is occurring?
- At what stage of product development should ethics review take place, and why does late-stage review fail?
- Name three failure modes in ethics review board design and what each leads to.