Put to: Greg Duff and Cindy Estis Green.
Why it matters to buyers: HSMAI’s account of Duff’s 2026 legal discussion explicitly identifies risks associated with AI-generated content alongside contract and compliance issues. Use that as a prompt to examine evidence controls, not as evidence that a particular tool is reliable or unreliable. (americas.hsmai.org)
Draft framing of a good answer: Ask the expert to demonstrate a chain from source document to extracted input, calculation and investment conclusion. Require dates, page or cell references, metric definitions and clear labels for missing or assumed values.
For AI-assisted underwriting, test how the workflow handles conflicting documents, unreadable tables, gross-versus-net revenue, duplicated fees and obsolete forecasts. Require human approval of material interpretations and independently checked arithmetic.
For data rights, request agreement-specific treatment of permitted access, export formats, retention, confidentiality, security and transition assistance. Do not assume that operational access means unrestricted ownership or reuse rights.
Sharp follow-up: “If we terminate tomorrow, which records can we lawfully export, in what usable format, and who verifies completeness?”
Evidence to request: Reproducible sample analysis, data-rights matrix, security terms and a documented handover process.
Interview status: No expert response has been recorded.