AI flood hits game stores: what Steam, itch.io and Nexus Mods now require

What's going on

If you're about to release a game, an asset pack or a mod right now, you'll run into new hurdles. Within a matter of days, several major platforms responded to the wave of AI-generated uploads. In early November, itch.io says it deleted "a few terabytes of AI slop." "Slop" means mass-produced, low-effort AI output. Since November 3, Nexus Mods has required AI disclosures for every new mod. And on Steam, indie studios are waiting up to three weeks for approval.

So far there have only been scattered reports. We put the rules side by side and check what the data actually says about their impact. Our thesis: The platforms aren't banning AI. They're regulating it through disclosure requirements and spam rules. What's being policed isn't the tool. It's the volume.

Steam: Disclose what players see, then wait

Since January 2024, Valve has required developers to declare AI use in a questionnaire called the "Content Survey." In mid-January 2026, Valve clarified the form. According to the Steamworks documentation, it covers content that ships with the game and is "consumed" by players: art, sound, story and localization. Marketing material on the store page counts too. Pure productivity gains are explicitly excluded:

"Efficiency gains through the use of these tools are not the focus of this section."

So you don't need to disclose a coding assistant in your editor. Valve distinguishes between two kinds of AI content:

• Pre-Generated content is reviewed like any other content. It must not contain anything illegal or infringe anyone else's rights.
• Live-Generated content is created while the game is running, such as chatbot dialogue. Here, developers have to describe their "guardrails," the safeguards against illegal output. Players can report this kind of content through the Steam overlay. Valve doesn't allow live-generated adult sexual content at all.

What's new is the backlog. In a Steamworks post on November 3, Valve wrote of a "significant increase in submissions." Reviews currently take up to 12 business days instead of the usual 3 to 5. Valve is also adding new items to the pre-release checklist. Among them: the game must have been played via Steam for at least five minutes in the past seven days. Valve doesn't mention AI as the cause. Tom Giardino of the Steam business team explicitly rules out a ban:

"We don't want to be gatekeepers telling developers which tools they should or shouldn't use."

Small studios pay the price. According to Power Up Gaming, "Wheel Make It" from publisher 2 Left Thumbs was supposed to launch on October 12. It was rejected twice: once because reviewers couldn't find the local co-op mode, and once because the resolution couldn't be changed with a controller. It finally came out on November 1. Despite more than 11,000 wishlists, it started with just 15 concurrent players. The publisher blames a "tidal wave of AI slop" in the queue. Kotaku names other delayed titles, such as "Food Devils," which had been stuck in review since October 12.

itch.io: The spam is the violation, not the AI

itch.io is a small indie marketplace with free hosting. According to Kotaku, the number of game pages created per day doubled from just over 3,000 to more than 6,000 between August 3 and October 30. Downloads and purchases didn't grow with it. According to itch.io, the growth comes from first-time uploaders, not a handful of mass uploaders. Asked why, the platform replied only: "Take a guess."

According to Aftermath, PC Gamer and Kotaku, these rules now apply:

• Disclosure: Mandatory for game assets since November 2024. Without it, a page won't show up in search. Tabletop, books, comics and soundtracks have now been added. You can't create or edit these pages without an AI disclosure. According to Kotaku, that holds even if the AI parts were later removed.
• Filter: If you don't want to see AI content, you can now hide it in search results too. 80.lv describes this as a "No AI" filter. But the filter is only as good as the uploaders' disclosures.
• Spam rules: Projects "created predominantly by algorithms or AI with minimal human input" are banned. So are random content with no curated design and floods of near-identical variants. AI content is allowed if it adds "substantial value" and is disclosed.

According to Aftermath, itch.io isn't planning a blanket ban. What got deleted or pulled from search was whatever counted as spam. Aftermath also names the weak spot: if you don't disclose AI, you're unlikely to get caught. The author found plenty of obvious AI titles without labels among the newest releases.

Nexus Mods: Three labels and a burden of proof

The biggest mod platform goes into the most detail. According to the November 3 announcement, summarized by ComputerBase and Inven Global, there are three categories:

1. AI-Generated Content: Textures, voices, dialogue, translations, music or code come mostly from an AI.
2. AI-Assisted: AI handles only individual steps, such as upscaling (scaling textures up to a higher resolution) or cleaning up audio. If you pick this category, you have to state publicly which tools produced what.
3. AI Media: Thumbnails, videos or descriptions. Even a handmade mod with an AI cover image has to be labeled.

According to ComputerBase, moderators can request project files or version history when in doubt, and change the label themselves if needed. Existing mods have a grace period until January 2, 2027. After that, a tiered process kicks in: first a notice, then a warning, account restrictions or deletion.

Nexus also rejects a ban. Its reasoning: a ban would just push authors to hide their AI use. Here, too, the real fight is about volume. Over the past year, 20,000 uploads were quarantined for extra review, twelve times as many as the year before. Nexus is also testing caps so that high-volume uploaders don't dominate the new-mods list. Mods labeled "AI-Generated" received 2.13 percent of the Donation Points paid out this year, the rewards Nexus distributes to mod authors.

What about everyone else?

Epic Games Store: No AI labeling. According to VGC, Epic CEO Tim Sweeney called Steam's requirement "really irresponsible" in June, and an "AI scarlet letter," in other words a public mark of shame.

GOG: We couldn't find a public AI policy for games. After criticism of an AI-generated sale banner in January, GOG declined to comment on "internal processes," according to GamingOnLinux.

Roblox: Games with AI that players interact with, such as chat, voice, images or 3D, have to show visible notices. Persistent AI conversations with memory are only available behind the 18+ "Restricted" label.

Apple and Google: Neither requires AI labels for apps. Apple relies on its rules against copycats (4.1) and spam (4.3), which reject "indistinguishable" variants. Google's policy on AI content targets harmful output, not where content came from.

What the data says about impact

Several counts with different methods show how common AI disclosures have become:

According to Kotaku and the Reddit thread Kotaku relies on, that would mean roughly 2,000 AI-labeled games in October alone. Haro's count came to about 530 per month in mid-2026. Our take: if both numbers are right, the volume quadrupled within a few months. That would go a long way toward explaining Valve's backlog. But the link isn't proven.

On how this affects a game's success: in part one of its study, Game Oracle finds about 21 percent of Steam titles from 2025 carried an AI label. For part two, the company analyzed a cleaned sample of 9,879 games released between January and October 2025, excluding free-to-play games and spam. The result: comparable games with an AI disclosure received about 52.6 percent fewer reviews. In the first month, that meant an average of 4 instead of 7. The distinction matters: what was mainly measured was the number of reviews, a rough proxy for sales. The ratings themselves were only slightly lower for games with at least 100 reviews, at 84.6 versus 88.3 percent positive. Game Oracle itself cautions that budget, marketing and polish can't be cleanly factored out. AI may simply show up more often in cheap projects. An independent analysis of more than 53,000 Steam games paints a similar picture: AI-labeled games reach the 100-review mark at only about 55 percent of the rate of other games.

What you should do now

• Steam: Disclose everything players see or hear, including AI art on your store page. Plan at least three weeks of buffer before release, because every rejection restarts the review. Play your build via Steam for five minutes beforehand.
• itch.io: Set the AI disclosure on tabletop, comics, books and soundtracks too. Don't upload near-identical variants.
• Nexus Mods: Review your old mods before January 2. Don't forget the thumbnails. If you choose "AI-Assisted," name your tools and keep your project files.

Our verdict

The thesis holds. None of the major platforms bans generative AI. itch.io deletes what counts as spam. Nexus fights back with quarantine and upload caps, Valve with checklists, and Apple has had its spam rule for years. Disclosure itself is the real penalty: it leaves the judgment to customers, and the Game Oracle data suggests customers are indeed judging.

That creates a contradiction. Honesty costs reach, and almost everywhere, enforcement relies only on self-disclosure and user reports. If you cheat, you have a good chance of getting away with it. That's exactly what Sweeney's criticism targets. On top of that come new problems no label can fix. The threat of AI-cloned demos has already pushed studios to pull their Steam demos. And Anthropic has barred Claude Code from recreating existing games. Our take: the next round will be less about labels and more about capacity. The winners will likely be the platforms that can filter out the flood most cheaply without punishing honest small studios along the way. Steam's backlog shows nobody has cracked that yet.