RESPONSIBLE AI, COMPLIANCE, AND RISK

LEARNING FROM NEAR MISSES Not every problem becomes an incident. A near miss is an event that could have caused harm but was caught in time, such as a safety check that blocked a bad action. Record near misses and review them, because they show where the safeguards worked and where luck may have helped. Teams that study near misses fix weaknesses before they cause real damage. EXAMPLES Example 1: Matched-pair bias test A hiring agent is tested with pairs of resumes that differ only in the applicant's name. If the scores differ in a consistent way, the team finds and fixes the cause before deployment. Example 2: Privacy by design An agent handling customer data never writes a full government identification number to a log. It stores only a masked partial value for audit purposes. Example 3: Auditability A loan agent keeps a tamper-evident record of each decision and the data it considered. The record is kept for the legally required period in case a regulator asks questions. Example 4: Risk classification A team rates a new feature as high risk because it affects people's jobs and is hard to undo. High-risk features require extra human review and sign-off before launch, in the spirit of the NIST framework. PRACTICE LAB CONNECTION Run a bias and fairness self-check on the Resume Review Agent lab's matching logic across Chapters 68 and 69. Test the prompts below against its outputs.