Harmful Outputs

Harmful outputs refers to content or actions an AI system produces that can injure people, mislead them, or break the law, such as dangerous medical or financial advice, discriminatory decisions, defamatory statements, or instructions for wrongdoing. For enterprises, these failures create direct legal exposure, consumer harm, and reputational damage, which is why regulators increasingly expect companies to test for them, monitor deployed systems, and document how they respond when problems appear. Governance programs typically address this risk through pre-release evaluations, content filters, human review of high-stakes outputs, and clear accountability for what the system is allowed to say or do.

1 item