AI Bias

Term: AI Bias

AI Bias is systematic differential treatment by an AI system across groups, whether those groups are defined by protected characteristics, operational context or data segments. It is the failure mode that fairness controls exist to prevent.

Why it matters

Bias is a significant AI-specific risk for UK organisations, and it carries direct legal weight. Under the Equality Act 2010, biased AI outcomes may create discrimination exposure for your organisation regardless of whether the bias was intentional, including where it arises from training data alone. Where an organisation places AI systems on the EU market or its systems affect EU persons, EU AI Act Article 10 may also impose data governance obligations on high-risk systems.

For boards and CISOs, this reframes bias from an abstract ethical concern into a measurable governance obligation. An AI system that produces uneven outcomes across groups is a compliance exposure, not just a technical defect.

How it works in practice

We treat Bias as the failure mode that the Fairness dimension of our CIA+EFT Framework controls against. AI Fairness is the property: consistent, equitable outputs across groups. AI Bias is the absence of that property. Fairness measurement and monitoring exist to detect bias before it reaches the people an AI system affects, whether in recruitment screening, fraud detection or access decisions.

This content is for informational purposes only and does not constitute legal advice.

Related terms: AI Fairness and the CIA+EFT Framework.