Kotaku says Epic Games and Microsoft have "suddenly" started treating AI as theft, because AI-built browser clones of their games are now appearing, even as both companies push AI inside their own studios; Epic CEO Tim Sweeney replied that the story "conflates a number of different things." The column ran on October 26, 2026. This post lays out what is reported, what each side says, and what the law has and has not settled, with claims attributed to the outlet that made them.
The conflict is worth understanding because it is the same dispute AI labs face from the other direction. Companies that use AI-assisted tools to build faster now find that those tools make it cheap for anyone to reproduce their products. How they answer shows where the line between "AI-assisted creation" and "unauthorized reproduction" is likely to be drawn. We covered the corporate side earlier in Epic's AI-first order for its Disney shooter and Epic's vibe-coded launcher.
Quick reference: reported versus established
| Item | Status |
|---|---|
| Epic filed a DMCA notice on a browser Fortnite recreation on Oct 10 | Reported by Kotaku, repeated by Tech Times and Gadget Review |
| Microsoft pursued takedowns on browser versions of Minecraft, Call of Duty, Halo | Reported by Kotaku; counts not verified |
| Open-source versions of Slay the Spire 2, Celeste, RimWorld appear in GitHub records | Reported; filer not established |
| Sweeney: story "conflates a number of different things" | His post on X, October 27, as quoted by outlets |
| Epic's terms restrict AI training on Unreal Engine and MetaHuman outputs | Acknowledged by Sweeney, who says Epic is working to remove it |
| AI-generated assets in Call of Duty | Claimed by Kotaku; needs independent verification |
| Hundreds of DMCA notices overall | Tech Times attributes this to Kotaku; unverified |
What exactly is being taken down?
The targets are AI-assisted recreations: browser-based ports and rebuilds made with coding agents. Tech Times summarizes Kotaku's report: developers are using AI coding tools to create browser ports and rebuild games for hardware they were not designed to run on, and "hundreds" of DMCA notices have reportedly been filed in response. Gadget Review adds that the number and current status of individual notices have not been independently verified.
The named cases are a browser-based Fortnite recreation (Epic, October 10) and browser versions of Minecraft, Call of Duty and Halo (Microsoft, over the previous month). Kotaku also points to GitHub's copyright records for open-source versions of Slay the Spire 2, Celeste and RimWorld, whose originals contain no generative AI content. The sources we read do not say who filed those indie-game notices or whether the developers authorized the projects, so do not assume Epic or Microsoft filed them.
This is the same trend we described in Xbox Labs and AI game decompiling: Microsoft's gaming chief acknowledged that AI-assisted recreations can challenge business models while making games more accessible, and said Xbox would be "responsible stewards of IP."
Copyright claim stamp over a page, representing DMCA takedowns of AI-built game recreations
What does Tim Sweeney say?
Sweeney posted a breakdown on X on October 27. The Gadget Review summary of his three points:
- Epic does not allow people to use Fortnite IP without permission or compensation, so a clone is not allowed.
- Epic's current terms restrict AI training on outputs from Unreal Engine and MetaHuman, which he acknowledged, adding the company is working to remove that restriction.
- He disputed that all AI training amounts to theft, pointing to a court ruling that training itself was fair use, which Futurism reports he used to defend Epic's internal use of Claude.
The first point is ordinary copyright: copying a game's characters, assets or code is infringement regardless of how the copy was made. The third is where the argument sits. Whether AI training is lawful is a different question from whether a finished clone infringes, and Sweeney's reply is that critics blur them.
What has the law settled?
Less than either side implies. In Bartz v. Anthropic, according to the summaries Gadget Review and Tech Times cite, a federal court held that using copyrighted books to train large language models was "quintessentially transformative" and fair use on the facts before it, while separately holding that downloading and retaining pirated copies was not fair use. Anthropic later settled with authors for a reported $1.5 billion, with destruction of files obtained from pirate sources, but it was not required to delete or retrain its models.
Three points follow. The ruling is fact-specific and did not create a universal safe harbor. Training and output are distinct questions: a model can be lawfully trained and still produce an infringing copy. And the case does not show that Epic's reported use of Claude infringes anything, as Tech Times notes. For how this fits the wider litigation, see our AI lawsuit tracker and the Stanford professors and Claude training post.
Where does "theft" come from?
The word cuts two ways. Futurism and Kotaku quote an unsealed filing in the New York Times case against OpenAI and Microsoft, in which Microsoft director of applied science Brent Hecht reportedly warned privately that creators could see AI models using their work as "an astonishing theft of unprecedented proportions," with another passage calling the AI buildout the "largest theft of labor in human history." That is a leaked internal comment, not Microsoft's public position. Another internal document, per Futurism, warned that generative AI could "significantly disrupt the employment of the very people who generated the data on which the foundation model was trained."
Epic's side is different. Tech Times notes Epic's terms prohibit using code or content extracted from its licensed products as training input for generative AI, including via prompts, and Sweeney criticized AI companies' use of pirated data in a June 2026 interview. Tech Times adds that these facts do not establish that either company breached copyright through its own AI activities.
Game controller with a green spark representing game developers weighing AI tools against intellectual property protection
Is it hypocrisy?
Here are the arguments, as reported.
The hypocrisy case (Kotaku, Futurism). Epic told its developers to use more AI to finish the Disney collaboration, and Microsoft's Activision is reported to use AI in art for recent Call of Duty games. Yet both file takedowns against AI-built clones. The critics' point is that the companies want AI to speed their own work and legal protection from others' AI.
The consistency case (Sweeney, implied by the law). Copyright protects works, not the tools used to make them. A company can use AI to write code and still enforce its rights against a copy of its game. On that view, there is no contradiction: infringement is judged by what was copied, not by whether a model or a person did it.
Both can be true. The practical issue is the one Gadget Review raises: large publishers have legal teams to monitor repositories and send notices, while the indie studios behind games like Celeste and Slay the Spire 2 typically do not. As AI lowers the cost of reproduction, enforcement capacity may matter more than the legal rule.
Where can you read the sources?
The original is Kotaku's column, Now That The Shoe Is On The Other Foot, Microsoft And Epic Games Suddenly Agree That AI Is Theft. Secondary coverage includes Tech Times, Gadget Review and Futurism. Sweeney's reply is on his X account, which the outlets quote; Gadget Review notes it did not independently verify the post's canonical URL. The takedown notices themselves are the primary evidence: GitHub publishes DMCA notices in a public repository, so you can check who filed which notice and when, rather than relying on counts in a column.
One more caution on framing. "Suddenly agree that AI is theft" is Kotaku's headline, an interpretation. Neither company has said AI is theft, and Sweeney explicitly rejects the idea. The factual core is narrower: notices were filed against AI-assisted recreations, and both companies use AI elsewhere. Whether that adds up to inconsistency depends on whether you think the law should treat a tool-assisted copy differently from a hand-made one, which it currently does not.
What should builders take from it?
- Do not clone shipped games. AI makes a port easy, but the code, art and characters are still someone's property. Browser ports of commercial games invite DMCA notices.
- Read engine and platform terms. Epic's EULA restricts AI training on Unreal Engine and MetaHuman outputs, per Sweeney. If you train on engine output, check the current terms and any announced changes.
- Separate training from output. A lawful training run does not make an infringing output lawful.
- Keep records. If you build with agents, log what you prompted and what sources you used, so you can show originality.
- Watch for the numbers. The counts of notices and the Call of Duty asset claim are unverified; wait for GitHub's records or company statements.
Related reading
Details reflect Kotaku, Futurism, Tech Times and Gadget Review reporting as of October 28, 2026; counts and some claims are unverified.