Mandatory marks put the cost on those who comply; those who mean to evade can go around

What it is

A newsroom’s generated image carries visible words and a watermark: the layout looks worse, and the platform may downrank it. The next account uses a tool that never marked, or washes the image, and carries nothing. The costs of a mandate — looks, reach, engineering — land on people willing to comply; people intending to deceive were going around anyway. Burden on the compliant means making marks an obligation of those who comply, without symmetrically punishing evasion, so the labeling system selects “honest synthesis” and lets “dishonest synthesis” through.

Pollution from missing labels is an environmental effect. This entry governs who pays for the mark, and who rides free.

Why it happens

If a mandate binds only cooperating platforms’ exits, evasion is switch tool, wash, redraw by hand, an offshore model. The compliant path has none of those exits, so the cost gap becomes a competitiveness gap: marked content looks more “fake” and is clicked less; unmarked looks more “real.” When recipients also read a missed detection as human, evasion is rewarded twice.

A design that only asks “did our users mark” turns the problem into a completion rate. The real distribution is “how much of the synthetic material in the environment is marked” — the denominator includes everything that never used your exit.

Studying it

Compare: mandate only this platform’s exit versus mandate plus distribution friction on unmarked content (downrank, “origin unmarked”). Dependent variables: reach of compliant content, reach of evading content, rate of recipients taking unmarked as human. Independent variables: aesthetic cost of the visible mark, whether the platform treats unmarked the same.

Measure whether evasion is still more readily taken as a record. If yes, the mandate is punishing compliance.

Where it stops holding

In a closed enterprise with one generation exit, a mandate can cover everyone and evasion paths are narrow. When the visible mark is extremely light and the watermark fully invisible, aesthetic cost drops and burden is mostly engineering — still heavy for small tools. Treating all unmarked as generated will injure human work; “to punish evasion” does not buy that. This entry does not treat the verification entrance, nor the semantics of a negative.

Applying it

• A mandate must come with equal or stronger distribution friction on unmarked content, or do not claim “the environment is labelled.”
• Lower compliance cost: visible mark optional with watermark kept, badge placements that do not wreck composition, no downrank of compliant content by the detector.
• Do not tell the public that “100% marked on this platform” means “unmarked things you see are human.”
• Check: pair a compliant marked image with an evading unmarked one in recommendation and in user choice. If unmarked wins on “more real” and marked loses on “ugly,” the burden is already on those who complied.

Related

• Same group: L3.10.1 Labels split into visible marks and embedded watermarks, which fail in different ways and fit different scenes · L3.10.2 Screenshots, paraphrase, and re-edits strip labels; survival falls with each hop · L3.10.3 A missed detection is not proof the content was not generated; lack of a mark is not evidence of human authorship · L3.10.4 A mark works only when someone verifies, so a usable verification entrance must come with it
• Nearby: L3.04 Content Labeling · L3.05 Copyright and Material Provenance
• Search terms: burden on the compliant · labeling evasion · asymmetric disclosure cost