A photo contest bans AI images; a platform just labels them

The process wall

Three documents, three different jobs for a disclosure label. World Press Photo's entry rules take the strictest position of the three: “AI-generated images are not photography... No synthetic or artificially generated images are allowed, and no use of artificially generative fill is allowed in post-production. Any use of these tools will automatically disqualify the entry.” The rules do carve out a narrower allowance for “smart tools, or AI-powered enhancement tools” such as denoise or automatic colour adjustment, but only where they do not add, remove, or alter information the camera captured; AI-powered enlarging tools are named as an automatic breach.

What the documents show

Content Credentials, the public-facing name for the C2PA provenance standard backed by more than 500 companies including Adobe, Microsoft, OpenAI and Google, takes a different approach: rather than banning generated content, it attaches tamper-evident metadata recording “the method of creation” and “a record of editing history” to an asset, so a viewer can check rather than simply trust. YouTube's disclosure policy sits between the two: it requires creators to disclose, via a setting at upload, any content that uses AI to “meaningfully alter or generate photorealistic content,” naming examples such as making a real person appear to do something they did not do, or generating a realistic scene that did not occur, while explicitly exempting “minor edits” like beauty filters and colour adjustments.

Authorship and limitation

The three documents are answering different questions with the same word. World Press Photo's rule is a categorical exclusion tied to what a camera captured, not a disclosure requirement at all — there is no disclosed-and-permitted category for a generated image in the contest. YouTube's policy is a disclosure requirement triggered by a threshold of alteration, applied by the creator at upload and enforced by the platform. Content Credentials is neither a ban nor a mandatory disclosure; it is a record a creator can choose to attach, which a viewer can choose to check.

What survives

This is an editorial checklist built from three separate documents, not a single standard: know which regime a piece of work is entering before assuming a label is sufficient or even permitted. A contest's own rules may make disclosure irrelevant by excluding the material outright; a platform's policy may require an affirmative disclosure at upload regardless of any attached metadata; and a Content Credential is a durable record that survives outside any one platform's rules, which a ban or a disclosure setting does not.

• Read a contest's entry rules for whether generated or AI-assisted work is disclosed-and-permitted, or excluded outright.
• Check whether a platform's own disclosure policy names the specific alteration threshold that triggers a required label.
• Ask whether a Content Credential was attached at export, since it travels with the file independently of any platform's disclosure setting.

A single generated image can be simultaneously disqualified from one contest, required to carry a platform label under a different rule, and correctly carrying a Content Credential that neither rule consulted; the documents describe three separate systems, not one shared standard.