What ASCI's AI Labelling Rules Change In The Edit
Read ASCI's new guidance as a disclosure requirement and you will implement it too late. AI labelling is decided by choices made in the edit, weeks before anyone senior looks at the file.
The guidelines sort AI use into three groups: what must be labelled, what needs nothing, and what no label can rescue. Only the middle group is about paperwork. The other two are about how a thing was made.
Where The AI Labelling Trigger Sits
A synthetic ambassador, a replicated voice, a fabricated setting that changes how the product is understood, a demonstration of something the product does not do. Each of those is a decision taken by a person choosing an approach, usually early, usually on grounds of cost or speed.
By the time a file reaches review, that decision is expensive to reverse. Which means compliance has to move upstream into the brief and the treatment, and the people who need to know the categories are the ones making the work.
Colour correction needs no label. A location that was never real does. Somebody has to know the difference while there is still time to choose.
The Group Nobody Is Reading Carefully
The prohibited list is the consequential one. Fabricated testimonials, synthetic exaggeration of results, unauthorised likeness. Those are not disclosure failures that a line of type fixes; they are simply not permitted, and an agency that treats the guidance as a labelling exercise will discover this late.
What To Put In Place Now
One question added to the treatment approval: does anything here fall in the label group or the prohibited group. It costs a minute and it moves the decision to where it can still be changed. ASCI publishes the full guidance at ascionline.in, and it takes effect three months from publication.
More from The AI Shift.