Skip to content

Contestability and Appeals for AI Decisions

Giving people a way to question and challenge decisions made or supported by AI systems.

Editorial team 1 min read

When AI influences decisions about people — loans, benefits, jobs, content moderation — they need a way to challenge outcomes.

Why It Matters

  • AI systems make errors.
  • People may have context the system lacked.
  • Appeal routes are required by some regulations.
  • Contestability builds trust and surfaces problems.

Elements of a Good Process

  • Notice: tell people a decision was made and that AI was involved.
  • Explanation: give understandable main reasons.
  • Route to challenge: a clear, accessible way to appeal.
  • Human review: a person with authority to change the outcome.
  • Timeliness: decisions reviewed promptly.
  • Correction: fix errors in data and outcomes.

Learn From Appeals

Track appeals and outcomes. High overturn rates signal model or process problems.

Avoid Rubber-Stamping

Reviewers need time, information and authority to disagree with the system.

Design Early

Build contestability into systems from the start rather than adding it after complaints.

More in Responsible AI

All Responsible AI guides →
Responsible AI Guide · 2 min

What Is Responsible AI?

The principles behind responsible AI — fairness, transparency, privacy, safety, accountability — and how to turn them into practice.

Responsible AI 2 min read 30 Apr 2026

Responsible AI Guide · 2 min

Understanding Bias in AI Systems

Where bias in AI comes from — data, labels, design and deployment — and why removing sensitive attributes isn't enough.

Responsible AI 2 min read 29 Apr 2026

Responsible AI Guide · 2 min

Fairness Metrics for Machine Learning

Demographic parity, equal opportunity, equalised odds and calibration: what each measures and why they can't all be satisfied at once.

Responsible AI 2 min read 28 Apr 2026

Responsible AI Guide · 2 min

Privacy in AI Projects

How to handle personal data responsibly when building AI: minimisation, purpose limits, de-identification and the risks of models leaking data.

Responsible AI 2 min read 27 Apr 2026