Human accountability
AI may assist research, analysis, drafting, organization and workflow design. A qualified person remains responsible for evaluating context, reviewing material and making or approving consequential decisions.
Confidential and sensitive information
Confidential, privileged, health, personnel and other sensitive information should not be entered into public or unapproved AI systems. Tools, access, retention, contracts and permitted uses must be evaluated first.
Accuracy and validation
AI output may be incomplete, inaccurate, biased or outdated. Material facts, citations, calculations, legal propositions and operational recommendations require verification appropriate to their risk and intended use.
High-impact decisions
AI output should not be the sole basis for significant employment, health, education, legal, financial or eligibility decisions. Human review must be meaningful, informed and authorized to change the outcome.
Governance and vendors
Organizations should define approved uses, decision rights, escalation, monitoring, documentation and incident response. Providers should be evaluated for privacy, security, data use, intellectual property, reliability and regulatory implications.
Transparency and review
Material use of AI should be disclosed when context, law or trust requires it. Because technology and regulation change rapidly, controls and this policy require periodic review.