Output integrity is the privacy risk that an AI system produces harmful or false statements about real people. A fabricated personal fact can cause harm when professionals rely on it without verification. section 5.0 section 5.1
Human review must involve proper scrutiny: approving AI output without checking it is not meaningful human oversight. Automation bias can lead people to rely on hallucinated claims and make unfair decisions. section 5.1 section 5.1
False or harmful claims about people can include:
- Invented medical histories. section 5.1
- Fabricated criminal records. section 5.1
- Statements misattributed to a person. section 5.1
