Design defect

A defect originating in the model's architecture or training, producing systematically wrong or harmful outputs for categories of input. Distinguished from a manufacturing defect (the model executed correctly but produced a one-off error) and from a deployment defect (the model was correctly designed but incorrectly used).

A flaw in a product's fundamental design, present in every unit of that product, that makes it unreasonably dangerous even when manufactured and used as intended. Distinct from a manufacturing defect, which affects only some units due to an error in production.

A systematic flaw in the conception or training of an AI system that affects every instance of the product, as distinct from a defect in a single unit or deployment. Examples include training-data demographic gaps or objective-function misalignment.

A finding indicating the robot, as specified, will not perform its function safely, reliably, or economically, requiring an actual change to the design rather than better documentation or evidence.