1
Fairness
Test for unequal outcomes and fix them before people are affected.
Related: Bias · Testing and validation reports · Stage 5
Principles
Six ideas behind every framework.
Almost every rule and standard comes back to these six ideas.
2
Transparency
Say when AI is in use and how it reaches its outputs.
Related: Transparency · Notices and disclosures · Stage 6
3
Accountability
Name an owner for every system, with clear escalation paths when things go wrong.
Related: Accountability · AI policy and approvals · Stage 1
4
Privacy
Collect only the data you need and protect it across training and use.
Related: Data minimization · Data governance records · Stage 4
5
Safety and security
Stress-test for failure and misuse, and keep monitoring after launch.
Related: Robustness · Logs and monitoring records · Stage 5
6
Human oversight
Keep people able to review, override and stop automated decisions.
Related: Human oversight · Human oversight procedures · Stage 6
Other frameworks use similar sets. For example, the Microsoft Responsible AI Standard (as described in the supplied study notes) lists fairness, reliability and safety, privacy and security, inclusiveness, transparency and accountability. Inclusiveness, designing so people with different abilities can use a system, is worth adding to your own policy. These are summaries; read the primary sources.
Was this page useful? Tell us what is missing or wrong ↗ · Report a correction