Induced from Ben O’Brien’s intended-use lecture at the KENX AI in GxP East 2026 conference (2026-09-15-1611-ben-obrien-intended-use-statements-gxp-ai-lecture), 2026-09-15.
The same model, two sentences apart
“Categorizes and assigns severity to incoming deviations” reads as high risk. “Drafts preliminary deviation summaries for QA review” reads as low risk. They can describe the same underlying model; the second names where the human handoff sits, so quality can see the control instead of inferring an uncontrolled decision.
Wording moves the auditable surface area
Narrow intended use shrinks everything downstream: risk classification, testing strategy, monitoring complexity, and inspection scope. A rewording can also flip which metrics matter, shifting the implementation from precision-tuned to recall-tuned in one sentence. The claim from the regulatory-classification side agrees: intended use, not the marketing name, determines classification. This is the wording surface of The intended use statement is the master control lever for GxP AI; Bounded inputs make AI validation tractable is the same shrink applied to the data box; and Assurance must scale with autonomy is the same shrink applied to the output rung. For a vendor, In GxP markets, the MVP is the customer’s intended use means this sentence is the product.