Ohio State Fair poster contest: the AI disclosure rule broke
The Ohio State Fair will ban AI from its poster contest in 2027 after three years of allowing it if disclosed. Here is why honour-system rules fail, and how to write one that doesn't.
The Ohio State Fair poster contest is dropping AI. Its contest page says the rules will be reevaluated for 2027, "which will include prohibiting the use of AI." When the contest started in 2024, AI was allowed if you explained it in your application.
That is a three-year experiment in honour-system disclosure, quietly ending. I think the interesting part is not the outrage. It is the mechanism: the rule was fine, the enforcement never existed.
🔍 What the contest page actually says
I want to be precise here, because the headline going around is louder than the source. The page lists five 2026 winners and does not state the medium of any winning entry.
| Rule | 2026 contest |
|---|---|
| Entries received | 38 |
| Eligibility | Ohio resident, 18+, professional or amateur |
| Entries per person | One |
| Format | 24" × 36", 2:3 portrait, .JPG or .PDF |
| Medium | Must be disclosed by the artist |
| Grand champion prize | $1,000, ribbon, social recognition, fair access |
| AI status | Permitted since 2024 if explained; to be prohibited in 2027 |
First place went to Christin Billips of Westerville. Second through fifth went to artists from Columbus, Reynoldsburg, Lewis Center and Hilliard. Entries for the 2027 contest open in January 2027.
The organisers' own reasoning: they "recognize that the use of AI has changed greatly over the last few years in ways we didn't anticipate."
That sentence is the whole story. A rule written in 2024 assumed a tool that no longer exists in the same form.
📊 Why "just disclose it" was always going to fail
Disclosure rules work when three conditions hold. Break any one and the rule becomes decoration.
- The declaration is verifiable. You can check a receipt. You cannot check "I painted this" against a JPG.
- Lying is costly. With one entry per person and a $1,000 prize, the expected value of shading the truth is positive and the downside is a ribbon you never had.
- The categories are clean. "Medium: digital" covers Photoshop, Procreate, a Stable Diffusion output, and a Stable Diffusion output repainted for six hours. Those are not the same thing, and the form has one box.
The third one is what actually killed it. Most real workflows in 2026 are hybrids. If I generate a background, composite it, hand-letter the type and colour-grade the result, what is the medium? Any honest answer is a paragraph, and forms do not accept paragraphs.
Key takeaway: A disclosure rule is not a policy. It is a question. If you cannot verify the answer and you cannot punish a wrong one, you have not restricted anything — you have only recorded what honest people told you.
🛠️ What detection actually buys you
The obvious fix is "we'll just detect it." I build image tooling, so let me be blunt about what each layer is worth in a contest where $1,000 is on the line.
| Method | What it catches | How it fails |
|---|---|---|
| EXIF / metadata check | Files straight out of a generator with tags intact | One re-save in any editor strips it |
| C2PA content credentials | Signed provenance from participating tools | Only if the whole chain supports it; screenshots break it |
| Statistical AI-image detectors | Obvious unedited generations | False positives on heavy retouching; degrades fast on new models |
| Requiring layered source files (PSD/PROCREATE) | Almost everything | Real friction for entrants; excludes traditional media |
| Requiring work-in-progress evidence | Almost everything | Cheap to ask for, annoying to fake |
You can look at the first two yourself in about ten seconds. Our EXIF metadata viewer reads what a file still carries, and the AI image detector gives you a probability, not a verdict.
Treat that probability as a reason to ask a question, never as evidence. A detector that is right 90% of the time will still wrongly accuse roughly four people out of 38 entries. In a small community contest, four false accusations is worse than one undetected generation.
The two rows at the bottom of that table are the ones that work, and neither is technical. Process evidence beats artefact analysis. Ask for layers, timelapse, or sketches. Someone who did the work has them lying around; someone who did not has to manufacture a plausible history, which is far more effort than entering honestly.
💡 How to write a rule that survives contact
If you run a hackathon at a Sri Lankan university, a design contest for a client, or a freelance marketplace category, you will face this exact decision within a year. Some rules I would actually write:
- Say what you are protecting, not what you are banning. "This contest exists to commission original human illustration" survives a model release. "No Midjourney" does not.
- Ask for process, not confession. Require source files or working stages at submission, from everyone. It is a uniform burden and it is checkable.
- Separate the categories instead of policing one. An AI-assisted track and a hand-drawn track both get judged, and nobody has to lie to compete.
- Publish the rule change before entries open, not after judging. Ohio is doing this correctly: the ban lands for 2027, entries open January 2027. Changing rules retroactively is how you end up with a lawsuit instead of a poster.
- Decide who owns the output. If AI is permitted, copyright on a purely generated image is contested in most jurisdictions, which matters when you print 5,000 posters. The AI content ownership checker is a starting point, not legal advice.
That last one is underrated. A contest is buying usage rights. If the winning file cannot be reliably owned, the sponsor has paid $1,000 for something they may not be able to defend.
🌐 What this means for you
If you are a designer or student here competing for international briefs, the direction of travel is now clear: the 2027 rules will be stricter than the 2026 rules almost everywhere. Start keeping your process artefacts now. Save layered files. Keep the ugly first drafts. That folder is about to become the thing that proves authorship, and it costs you nothing to maintain.
If you run any kind of submission process, do not spend your budget on detection. Spend it on requirements that are cheap for honest participants and expensive for dishonest ones. Ohio State Fair ran a permissive rule for three years, watched the ground shift, and is changing it in the open with a full contest cycle of notice. That is a reasonable way to be wrong.
The failure was never allowing AI. It was writing a rule whose only enforcement was a text box.
Original source
AI poster wins Ohio State Fair contest