itch.io AI Game Policies in 2026: A Disclosure Workflow for Indie Devs

Why itch.io's AI stance matters more in 2026 than it did in 2023

When itch.io published its AI disclosure guidance, it took a position that was unusual for a major storefront: it did not prohibit AI-generated content. Instead, it asked creators to be honest about how AI was used. Valve went the other direction on Steam, requiring disclosure during submission and moderating AI content on a case-by-case basis, while Apple's App Store review guidelines now ask for disclosure of generative content in certain contexts. itch.io sits in a third category — permissive but loud about labeling.

That distinction matters because itch.io remains the default first stop for prototypes, game jams, and experimental work. In 2025 the platform ran its usual slate of jams — the recurring Game Off and the long-running Ludum Dare partnership — and generative tools were already embedded in many entrants' pipelines. By 2026, a large share of small-team projects touch AI at some stage: a concept pass through Midjourney, a music bed from Suno, voice lines from Play.ht, or a 3D prop from Tripo3D. The policy question is no longer "is AI allowed?" It is "what exactly did you use, and did you say so?"

What itch.io actually asks for

Disclosure, not prohibition

The core of the itch.io position is that games using AI-generated assets should say so on the project page. The platform has framed this as a matter of audience trust rather than a technical restriction: buyers and players should know what they are downloading. In practice this means a clearly visible note in the description or a dedicated section, not a buried line in a changelog.

The practical ambiguity is that itch.io does not publish a rigid, itemized taxonomy the way some storefronts do. There is no mandated checkbox for "AI-assisted" versus "AI-generated." That flexibility cuts both ways: it lets a solo developer describe a nuanced pipeline honestly, and it lets a bad-faith actor write a vague sentence and move on. If your project is reported, the burden falls on you to show that your disclosure was clear and proportionate.

The classification problem nobody solves cleanly

Most disputes about AI disclosure are really disputes about categories. Consider four cases a typical indie project might contain:

• Fully generated assets. A background image produced end-to-end by an image model. Easy to label, easy to defend.
• AI-assisted assets. A human-painted sprite refined with an AI cleanup pass, or a base mesh generated in 3D AI Studio and then retopologized by hand in Blender. This is where most teams live, and where most disclosure language is too vague to be useful.
• AI in tooling only. Autocomplete in your code editor, an AI-assisted bake in your DCC tool. This is generally not what players care about, but saying nothing can look evasive if your page otherwise leans on AI marketing.
• AI in runtime. Procedural dialogue, generated NPC behavior, live asset synthesis. This is a design feature, not a production shortcut, and it deserves its own sentence.

A useful test: if a player would feel misled upon learning the fact, disclose it. That standard is stricter than itch.io's written guidance and safer in a report scenario. For teams producing music, the same logic applies — an AIVA or Soundraw bed used as a placeholder that ships in the final build is a shipped AI asset, not a placeholder.

Real cases and concrete signals

Steam's disclosure flow. Valve's AI disclosure requirement, introduced in 2024 and still evolving into 2026, forces developers to describe both pre-generated and live-generated content at submission. It is a useful comparison point because it demonstrates that disclosure-first regimes are workable at scale — and because many itch.io projects later migrate to Steam, where the same asset list must be re-declared. Build your asset inventory once and reuse it.

The 2024–2025 asset-marketplace turmoil. Epic's decision to allow AI-generated content on Fab, and the resulting backlash and policy clarifications, showed how quickly community sentiment can force platform rule changes. itch.io's guidance should be read as a pre-emptive move in the same direction: get ahead of the argument by making labeling normal.

Jam-scale data. Ludum Dare entries routinely number in the thousands per event, and post-2023 editions saw visible debate in comment threads about AI-generated art and audio. The practical lesson for a jam participant in 2026 is that jam communities often enforce stricter norms than the host platform. Disclosing on your itch.io page costs you nothing; getting called out in a jam community costs you reputation.

Tooling maturity. By 2026, AI asset tools have moved from novelty to pipeline fixtures. A generated character base from Kaedim or CSM with PBR materials, animated via Cascadeur or DeepMotion, scored with Udio, and voiced with Resemble AI is a plausible solo-dev stack. Each of those steps produces a labelable artifact. The policy burden scales with the size of your AI surface area.

A practical disclosure workflow

1. Keep an asset ledger from day one. A single spreadsheet with columns for asset name, tool used, generation method (full / assisted / tooling-only), and whether it ships. Retrofitting this after a report is painful.
2. Write disclosure at the asset-class level, not the vibe level. "Background art generated with Midjourney, then repainted in Krita" beats "some assets were AI-assisted."
3. Separate production AI from runtime AI. If your game generates content live, say so explicitly and describe what the player will see. Reviewers and players treat these as different claims.
4. Name the tools where it helps. Naming Photoshop (AI) or Suno signals a specific, auditable workflow. Vagueness reads as concealment.
5. Put the disclosure where it is seen. A short section near the top of the description, plus a line in the credits. Do not rely on a devlog post from eight months ago.
6. Re-check before major updates. If you add generated voice or music in a patch, update the page. Stale disclosures are the most common failure mode.
7. Handle 3D and animation carefully. A model generated with Tripo3D and cleaned in Blender is still a generated asset. The cleanup does not erase the origin, and claiming otherwise invites disputes.
8. When in doubt, over-disclose. The downside of an extra sentence is negligible. The downside of a credible accusation of concealment is a takedown or a community pile-on.

Outlook

itch.io's approach will likely stay permissive, because its identity depends on being the low-friction place to publish. The pressure point is enforcement: as more AI-heavy projects appear, the platform may add structured fields to the upload flow — a checkbox and a free-text box would resolve most ambiguity without banning anything. Watch for that change, because it would turn today's judgment call into a form you must fill out correctly.

For developers, the strategic takeaway is boring but durable: treat disclosure as part of your build pipeline, not as a marketing decision. Document your tools, write specific labels, and keep the ledger current. That habit survives whichever way itch.io, Steam, or the next storefront tightens its rules — and it means a policy shift costs you an afternoon, not a launch.