Epic's CEO takes aim at Steam's AI labels as AI-made guides and game copies spread

Welcome to the first long read from The Daily Token Gaming. Tuesday, October 27, had no single blockbuster announcement in AI and games. Instead, the day was full of arguments about the same question: what happens when AI makes it cheap to produce things, from whole games to copies of other people's games to the guides you look up when you're stuck?

The biggest of those arguments is about labels. Epic Games' CEO says the AI disclosures on Steam are political, while the latest Steam Next Fest suggests players aren't flocking to games that use AI. Elsewhere, an AI game-making app launched on desktop, studios filed takedowns against AI-assisted copies of their games, AI-made guidebooks and fan wikis are misleading players, and some of Japan's best-known developers explained where they think AI belongs.

Tim Sweeney takes aim at Steam's AI labels

Steam, the biggest store for PC games, asks developers to say on their store page when a game uses generative AI. That's the kind of AI that creates new material, like images, voices, text or code, from a simple prompt. Players can then decide for themselves whether that matters to them.

Tim Sweeney, the CEO of Epic Games, which runs the rival Epic Games Store, has been attacking those labels in a string of posts on X, GamesIndustry.biz reports; the first went up late last week. He argued that the main use of the disclosures is to drive cancel campaigns against games and cost developers revenue, and wrote that "blocking free choice to adopt productive technology is a political act." He also compared the labels to forcing developers to write political essays or list their voting history.

In the same thread, Rock Paper Shotgun notes, Sweeney drew a line between AI labels and age ratings, which he said help parents make informed decisions. He also argued that Valve, which already takes a 30% cut of sales on Steam, is costing developers even more by fueling what he called AI hate campaigns. When someone pointed out that Fab, Epic's own marketplace for game assets, asks sellers to tag content made with generative AI, Sweeney said that was different because Fab is a platform where businesses sell to other businesses.

Not everyone buys the argument. Rock Paper Shotgun's view is that a disclosure is simply information for buyers, much like the label on a food package, and that it doesn't force anyone to write a political statement. If players react badly to what a label tells them, the site argues, that's a signal about what customers want.

This isn't a new fight. Back in June, Sweeney said it was "really irresponsible" of Valve to label games that use AI, according to GamesIndustry.biz. Epic itself is moving fast in the other direction: it recently said it "vibe coded" its Epic Games Launcher as part of a broader push into AI, and Sweeney says Epic has been extremely open about how it uses the technology. In his latest jab, GamesRadar+ reports, he argued that Valve has held gaming back for years by locking PC games into Steam's own social features.

What Next Fest says about AI games

So do players actually care whether a game uses AI? A small data point arrived this week. Steam Next Fest is Valve's regular event where players can try free demos of upcoming games, and the October 2026 edition had more than 8,000 of them, PC Gamer reports.

Valve has now published the 50 most-played demos from the event. PC Gamer counted just 4 among them that declare AI use in development, and reads that as players by and large passing on games with obvious generative AI. The list is topped by Worst Class Ever, a party game for 1 to 6 players, and it includes 7 incremental games, the genre where numbers keep going up while you click or wait.

One of the 4 AI games is Teach My Little Sister How To Drive, which leans on a character powered by a large language model, the same kind of AI that runs chatbots. According to PC Gamer, that character has apparently been costing its developer more than $1,000 a day. It's a reminder that AI characters in games aren't just a design choice; they can be a real running cost.

A top-50 list is one snapshot rather than a survey, and plenty of things decide which demos people play. Still, it gives both sides of the labeling debate something concrete to point at.

Remix brings its AI game maker to desktop

While the label fight goes on, making games with AI keeps getting easier. Remix is an app where you describe the game you want and its AI writes the code. That's often called vibe coding: explaining what you want in plain words and letting the AI do the programming.

On Monday, Remix launched a desktop app for Windows, Mac and Linux, joining its web version and its iOS and Android apps, GamesBeat reports. The company says about 1.5 million people play and create on Remix, and that creators have made roughly 100,000 short games on mobile since it started there in 2025. Remix hosts the games and adds leaderboards, achievements and real-time multiplayer. Creators own what they make and can export the code to take a game elsewhere.

Playing and sharing games is free. Building with Remix's AI starts at $5 a month, and creators can also connect an existing subscription from Anthropic, OpenAI or Google. To pull in new players, Remix is running a scavenger hunt called The Missing Bitcoin, with 1 Bitcoin (or its cash equivalent) as the grand prize and a total prize pool of more than $100,000. Its final stage opens on November 12.

Remix's CEO, Charlie Blackstock, said in a statement that the company wanted to change who gets to make games, and that some of its most creative builders are kids who have never written a line of code. Remix raised a $5 million seed round in 2025 led by Archetype, bringing its total funding to $6.75 million. As with any company's user numbers, those figures are Remix's own.

AI is making it easier to copy games

The same tools that help people make games can also help them copy games. Kotaku reports that AI coding tools have made decompiling dramatically faster. Decompiling means taking a finished game, which ships in compiled form that people can't easily read, and turning it back into code they can read and change. According to Kotaku, projects that used to take months or even years are now appearing daily, as unofficial versions that run in a web browser or rebuilt PC versions, and companies have sent hundreds of takedown notices in response.

Some of those notices are public. GitHub, the site where programmers share code, publishes the copyright takedown requests it receives. This month they include a notice from the makers of Slay the Spire 2, dated October 6, which says the game's source code has never been published or licensed. A RimWorld notice from October 12 was processed against a whole network of 33 copies of the same project, because the original was still being copied when the notice arrived. And a Celeste notice from October 14 targets the game's files inside a browser port. Kotaku points out that none of these indie games use generative AI themselves.

Big companies are filing too. Kotaku says Epic sent a takedown over a browser version of Fortnite on October 10, and that Microsoft has gone after browser versions of Minecraft, Call of Duty and Halo over the past month. Kotaku's argument is that big publishers can keep sending lawyers after copies, while small studios that publish their own games are the ones who really get hurt.

AI is also shaking up console hacking. In a separate story, Kotaku describes a PlayStation 5 hacker who says Sony paid him $50,000 through its bug bounty program, a reward for finding and reporting security holes. The hole was a hypervisor exploit, a flaw in one of the deepest layers of the console's security. He says he only reported it after someone else had found the same bug using AI. Kotaku notes that because it was reported in mid-October, this particular hole won't help anyone pirate GTA 6 when the game comes out on November 19.

AI-made guidebooks and fan wikis are misleading players

If you've ever been stuck in a game, this one is for you. Automaton reports that searching Amazon Japan for Fire Emblem: Fortune's Weave, which Nintendo released for the Switch 2 on September 17, turns up at least 9 guidebooks, none of them official. Their covers look heavily AI-generated, the sample pages are walls of text with no images, and the 4 books that have reviews have only one-star ratings. Similar guides have appeared before for games like Slay the Spire 2, Dragon Quest VII Reimagined and Pokémon Pokopia.

Automaton points to Kindle Direct Publishing, Amazon's free self-publishing service, which prints physical books only when someone orders one. Combined with generative AI, that makes it cheap to churn out "books" with little risk. Amazon's rules require publishers to tell Amazon when they use AI, but Automaton couldn't find any rule that buyers must be told, and none of the 9 books showed a disclosure.

It isn't only books. Eurogamer found fan wikis that appear to be AI-made, listing companions and skills that don't exist and crowding out real community wikis in search results. They mostly target smaller indie games, where there may not be a big fan community to fill the gap. Tim Bender, CEO of the publisher Hooded Horse, which runs official wikis for its games, told Eurogamer his company has to play whack-a-mole with these sites. One operator behind a network of more than 20 game wikis admitted to using generative AI to make them. In Bender's words, "these tools have made it very easy for bad actors to flood things."

Real fan wikis are also getting hit by AI crawlers, the bots that collect web pages for AI companies. The volunteer-run Dune: Awakening Community Wiki logged 7 million requests from AI crawlers in 30 days, 5 million of them from Meta, Eurogamer reports, and its maintainer says server costs have gone up between 3 and 10 times because of generative AI. Eurogamer also notes that the big wiki host Fandom had to roll back its own AI features over inaccuracies. The simple advice: if a guide or a wiki feels off, check who's behind it.

Japan's game veterans on where AI fits

Finally, two interviews show how differently veteran Japanese developers see AI. Kotaro Uchikoshi, the creator of the Zero Escape series, announced a new studio, Crash Keys, on October 15. In an interview with the Japanese site Game*Spark, reported by Automaton, he described his own stance on generative AI as positive and said the industry shouldn't reject it on reflex.

His model is a split of roles: humans create the initial design and story, and AI helps bring them to life, which could shorten development. He added that this only works if the AI can actually reproduce the world the creator has in mind, rather than inventing things on its own. He also said a project his new studio had to cancel might have turned out differently if the team had been able to use AI.

Meanwhile, three veterans behind Final Fantasy and the Mana series, Yoshinori Kitase, Hiromichi Tanaka and Koichi Ishii, discussed AI with Famitsu for Mana's 35th anniversary, as reported by Automaton and GamesRadar+. Tanaka said future creators will need to master AI, and expects more indie games as AI changes programming, while flagging copyright concerns around AI-generated graphics. Kitase agreed that AI can make development more efficient, but said the first creative step, which he calls turning zero into one, must always come from people. Tanaka added that the industry will become even more of a competition of ideas. Ishii is excited about faster prototyping, but warned that if AI does everything on its own, "it could end up feeling like it's missing human warmth."

The big picture

If one thread runs through Tuesday, it's that AI keeps lowering the cost of making things. That's good news when it lets a kid build a game in an app like Remix, or helps a small team ship sooner. It's bad news when the same ease lets someone rebuild an indie game's code without permission, or fill Amazon and search results with guides nobody checked.

That's why the fight over labels matters. Sweeney sees Steam's AI disclosures as a tool for cancel campaigns; their supporters see them as basic information for buyers. Next Fest's most-played list doesn't settle the argument, but it does suggest that, for now, most of the demos players chose didn't disclose any AI use.

And the developers with the longest track records seem to agree on one thing, even when they disagree on the rest: AI can speed up the work, but the original idea still has to come from people. We'll be back tomorrow with the next one.