Fairness means treating relevant individuals or groups equitably and avoiding unjustified disparities in AI systems or their use. In practice, fairness is both a model risk and a use-and-governance responsibility: prompt design, response interpretation, policies, and organizational procedures all influence whether AI is used fairly. section 8.0 section 8.2.1
Build fairness safeguards around the system and its users:
- Include fairness alongside privacy, security, transparency, and accountability in a short, practical AI-use policy. section 8.2.1
- Involve diverse teams in implementing, using, and reviewing generative AI so different perspectives can reveal bias and support more inclusive use cases. section 8.5.3
- Set ethical guidelines for detecting and addressing bias, and require fair and transparent use with clear accountability. section 8.5.3
- Evaluate fairness throughout the model life cycle, not only during design; update models and approved use cases when new data or biases appear. section 8.5.4
A concise AI-use policy
A small organization planning an AI pilot can begin with a one-page policy stating its purpose, scope, legal and ethical commitments, and expectations for responsible use. The policy should identify who may access the AI, which applications and documents are covered, and how fairness, transparency, accountability, privacy, and security will be handled. Healthcare, finance, and public-service applications may require stricter controls. section 8.2.1
Fairness is not a one-time approval. Ongoing monitoring helps keep outputs and approved uses accurate, fair, and current as data changes and previously unnoticed biases are identified. section 8.5.4
