Who Audits AI Decisions? Why We Need Sector-Specific AI Ethics Boards

AI is already influencing decisions that affect people’s lives: which patient receives priority, which student is considered “at risk,” or which customer is shown a particular offer.

AI ethics therefore needs to move beyond statements of good intentions. It must become a concrete governance priority for companies and governments.

The risks are real

We have already seen the harm that AI and algorithmic decision-making systems can cause:

• In the United States, a widely used healthcare algorithm systematically underestimated the care needs of Black patients.
• In England, the statistical approach used to calculate exam grades in 2020 produced unfair outcomes for some students. Following public criticism, the grading approach was changed.
• Amazon abandoned an experimental résumé-screening tool after discovering that it disadvantaged female candidates.
• In the Netherlands, thousands of families were wrongly accused of fraud in the childcare benefits scandal. The process, which involved algorithmic risk assessments alongside institutional failures, had severe consequences and led to the government’s resignation.

An error in a system may not remain confined to a single decision. It can be repeated across many decisions as the system continues to operate.

One set of principles is not enough

The risks differ from sector to sector:

• Healthcare: Is the model equally reliable across ages and population groups? Can healthcare professionals challenge and override its recommendations?
• Education: Can students appeal an algorithmic assessment? Is children’s data protected?
• Retail: Is personalized pricing fair? Would customers accept a particular offer if they understood why it was shown to them?
• Finance, HR and public services: Can a rejection be explained and reconsidered?

We should draw lessons from ethics committees in clinical research and establish independent, sector-specific review structures for AI. Each board should understand the risks within its sector and the people affected by its decisions.

Ethics boards need real authority

A board that cannot stop a project risks becoming a structure that provides the appearance of ethical oversight without meaningful influence.

An effective board should:

• Have direct access to senior leadership and the authority to pause or stop a project under clearly defined conditions.
• Include legal and ethics experts, technical specialists and representatives of the people affected.
• Review bias testing and impact assessments before a system is deployed.
• Monitor deployed systems regularly.

What governments and companies should do

I believe governments should require independent ethical review in high-risk areas. Public-sector systems should be documented in transparent registers that protect personal data and security. People should have accessible ways to challenge automated decisions.

The EU AI Act provides an important framework for risk-based obligations and human oversight. Under certain conditions, it also applies to organizations outside the EU. The sector-specific ethics boards proposed here are a governance approach that could complement that framework.

Companies should begin by creating an inventory of their AI systems, assessing their risks and assigning accountable owners to each system. An AI management system standard such as ISO/IEC 42001 can help structure this work.

Accountability must be visible in practice

Ethics boards should help organizations develop and use AI with confidence.

That requires clear authority, ongoing monitoring and accountability alongside ethical principles. Organizations that establish these structures today will be better prepared for future regulation and public expectations.

Who audits the decisions AI influences in your organization? When a system causes harm, who has the authority to stop it?