Maybe the Future Is Not Detection, but Accountability

That can include:

sketches,
raw generations,
intermediate files,
Photoshop versions,
and documented approval processes.

That completely changes the question.
Instead of asking:

Can we prove from the final image that AI was used?

perhaps organisations should ask:

Can we prove how the image was created?

Those are two very different standards.

From Pixel Forensics to Process Evidence
A professional creative production chain could look something like this:

Original source
→ Reference material
→ Prompt / AI model
→ Raw generation
→ Human selection
→ Photoshop / Illustrator
→ Colour correction
→ Brand review
→ Rights check
→ Human approval
→ Publication

Suddenly Human in the Loop stops being a fashionable phrase.

It becomes part of the evidence trail.

My earlier examination already arrived at a simple rule:
AI creates humans check the company decides.
That sounds almost bureaucratic.

But once origin becomes harder to reconstruct later, documentation becomes part of creative responsibility.
Human in the Loop Is Not Clicking “Approve”
Human oversight only matters if the human actually understands what is being checked.

For a fashion company, that could mean verifying:

product colour;
material;
logos;
stitching;
shape;
anatomy;
rights;
source material;
misleading claims;
synthetic identities;
final disclosure requirements.

Interestingly, Zalando’s own current partner guidelines now contain exactly this type of concern.

AI-generated assets can be rejected when they distort products, create implausible environments, produce inconsistent models or fail to represent garments accurately.

Its current video guidelines likewise require realistic fabrics, colours, textures, proportions and model consistency.

That makes the original examination almost strangely prophetic.

The theoretical concern became an operating standard.
Authorship Makes Everything Even Messier

Now imagine:

AI generates the first image.
Then I spend four hours changing:
composition,
lighting,
colour,
typography,
structure,
facial details,
background elements,
layout,
and narrative.
The final work may look nothing like the original generation.

Where exactly is authorship located?
In the prompt?
In the first output?
In the selection?
In the manual changes?
In the concept?
Or in the combination of all of them?

The second WBS examination explicitly asks students to confront the level of human creative contribution required for copyright protection.

Technology prefers simple buttons:

Generate.
Edit.
Export.

Law prefers definitions.

Creative work increasingly lives somewhere inconveniently between them.

C2PA and the Dream of Perfect Provenance
One proposed answer is provenance.

Instead of analysing an image afterwards and asking:

Does this look artificial?

we attach information about its history.

Which tool created it?
Was it modified?
What was the source?
Which transformations occurred?

That is a much more sophisticated idea.
And Zalando’s current guidelines are already moving in this direction.

Its partner documentation recommends or requires forms of invisible marking and AI tags for several AI-production workflows, while distinguishing between hybrid imagery, background replacement, face swaps and AI models.

But provenance has one obvious enemy:

the normal creative workflow.
Files are:
copied,
cropped,
screen-captured,
compressed,
renamed,
re-exported,
downloaded,
combined,
uploaded,
and edited again.
Technology can preserve provenance.

It cannot guarantee that humans will preserve the chain.
Shadow AI May Be the Bigger Corporate Risk
The WBS examination also raises something far less visually exciting:

Shadow AI.

An employee may quietly use an unauthorised public AI service because it is convenient.

Into that service they might paste:
customer information,
commercial plans,
campaign drafts,
confidential documents,
source code,
or internal strategy.

The examination therefore asks companies to think about whitelists, approved enterprise tools, training and IT-security controls.

This may ultimately be a larger corporate risk than an incorrectly labelled Instagram picture.

A missing AI label is visible.

A confidential strategy pasted into the wrong AI system may not be.

Less glamorous.
Much more dangerous.
Zalando Is Now the Real Laboratory
What makes all of this particularly interesting is that Zalando is no longer simply an academic example.

It is actively building the environment these examinations describe.

AI-generated marketing content.
AI-generated product video.
Hybrid photography.
Digital twins.
AI-assisted product onboarding.
Personalisation.
Automated image processing.
And strict requirements about product realism and AI workflows.

So the original question:

Should Zalando use AI?
has become obsolete.

The better question is:

How do you govern AI when AI becomes ordinary infrastructure?

Maybe We Are Asking the Wrong Question
Perhaps the future will not contain a perfect detector that looks at every creative work and declares:

HUMAN
or
AI

Creative production is becoming too hybrid for that.
Photography already uses computational processing.
Photoshop includes generative tools.

Illustrator combines manual and automated creation.
Video editing uses AI masking, restoration, tracking and generation.

Designers generate concepts and rebuild them by hand.
Photographers apply AI denoise to real images.

Brands create AI video from real product photography.
Digital twins originate from real people.

At some point the binary classification becomes almost absurd.

Better Questions

Instead of asking only:

Was AI used?

perhaps professionals should ask:

What role did AI play?
What was genuinely photographed or created by a person?
What was generated?
Was the audience materially misled?
Was the actual product represented truthfully?
Were the source materials legally usable?
Were private and confidential data protected?
Did meaningful human review take place?
Can the production process be reconstructed?
Who ultimately accepted responsibility?

Those questions are harder, they are also far more useful, final Thought, we spent years asking whether AI could create convincing content.

That question has been answered.

Zalando shows that hybrid AI production is no longer experimental.

Guess and Vogue show what happens when a synthetic model becomes indistinguishable from conventional fashion advertising.

The Brad Pitt fraud shows what happens when synthetic identity becomes a tool for manipulation.

And modern generative video shows that even an event involving some of the most recognisable faces in the world no longer needs to have happened for us to see it.

The next question is therefore much harder:

When human creativity and artificial generation become part of the same production chain, who can still prove where one ended and the other began?

Maybe the future of AI governance will depend less on inspecting the final pixels.

And much more on proving the journey that created them.

Because pixels can be edited, metadata can disappear, Detectors can be wrong.

But a professional organisation should still be able to explain:

what it made,
how it made it,
and who was responsible.

The real challenge is not simply detecting AI.

It is proving the story behind what we create.

Diana Trinchinet