An AI Takedown Bot Pulled a 15-Year-Old Open Source Project Off Google Play. It Is the Second Time, and Last Time Google Took 46 Days to Undo It.
Luanti, the open source voxel game platform formerly known as Minetest, is currently not on the Google Play Store. It was removed on a DMCA notice filed by a brand-protection company called Tracer.AI on behalf of Microsoft, alleging that Luanti infringes Minecraft's copyright.
The notice cites one thing: US Reg. #TX 8-192-097, which is the copyright registration for Minecraft Java Edition 1.9. It does not name a single allegedly infringing asset. Luanti ships no games and no game assets at all. It is an engine plus a content browser, and the entire set of textures that comes with it fits in one screenshot on their blog.
The same company filed a substantively identical notice against the same app in March 2023. Luanti counter-noticed and won. Google restored the app 46 days later, against a statutory deadline of 10 to 14 business days.
That gap between 14 and 46 is the part of this story that should concern you, and it has almost nothing to do with voxel games.
The asymmetry is the entire problem
Section 512(g)(2)(c) of the DMCA requires a service provider to restore removed material "not less than 10, nor more than 14, business days following receipt of the counter notice." That is the law. Google took 46 days in 2023 for this exact app, filed by this exact company, over this exact claim.
Now price out both sides of that exchange. Tracer.AI markets itself on volume. Their own 2024 marketing claims "85% faster takedowns," review times "six times faster than traditional methods," "100% more reviews month-over-month," and "44% more takedowns month-over-month." Filing a notice is a workflow step in an automated system optimized for throughput.
Receiving one costs you your distribution channel for somewhere between two weeks and two months, plus whatever legal comfort you need before signing a counter-notice that exposes you to jurisdiction and perjury liability. Luanti is a non-profit with volunteers and it still describes the process as something they lack resources to keep fighting. Their words: "If we were a small company relying on income from the app, this would be especially devastating."
I want to be careful here about what is actually established. Microsoft and Mojang do own Minecraft's assets and do have the right to enforce that. What is at issue is a notice with no specificity, filed twice, by an automated system whose vendor advertises speed and volume as the product.
This is a pattern, not an incident
In February 2026, Tracer.AI filed a DMCA takedown on Microsoft's behalf against Allumeria, an indie game with a similar voxel art style. Allumeria came off the Steam store. Microsoft eventually dropped the notice after the story got enough public attention, which is a resolution mechanism, but not one you can plan around.
So that is three known filings from one vendor: Luanti in 2023, Allumeria in February 2026, Luanti again in August 2026. Two of the three were resolved in the developer's favor. The third is pending. None of them appear to have cost the filer anything.
The underlying legal point Luanti makes in their post is correct and worth restating, because it is the thing an automated similarity detector cannot evaluate. Copyright does not protect ideas, systems, or methods of operation, per 17 U.S.C. § 102(b). Cubes are not ownable. Minecraft itself was inspired by Infiniminer in 2009. Block games are a genre, and a genre is not a copyrightable work. A model that flags visual similarity has no way to represent that distinction, and a takedown pipeline optimized for throughput has no incentive to.
What you should take from this if you ship an app
You have this exposure if you distribute through any single store, and the specific flavor of it does not require you to be near a big company's genre. Any automated detection system with a false positive rate above zero, pointed at a catalog of millions, will produce false positives at volume. You are one of the millions.
Three things that are worth doing before you need them, rather than after:
Have a second distribution channel that already works. Luanti's Android app is on F-Droid and available as a direct APK from their own site. That is why this is an inconvenience for them rather than an extinction event. If your only path to users is one store listing, your business continuity plan is a support ticket. Set up the alternate channel while nothing is wrong, because the moment you need it is the moment you have no time to build it.
Document asset provenance now, in a file, in the repo. Luanti could respond immediately because they already had a LICENSE.txt attributing every font and asset, and they could point at the exact December 2023 release where Minetest Game stopped being bundled. If you had to prove tomorrow that every image, font, sound and icon in your app is yours or properly licensed, how long would that take? For most solo projects the honest answer is days, and you will be asked to do it under a clock.
Know your counter-notice posture before the notice arrives. A DMCA counter-notice requires a statement under penalty of perjury and consent to jurisdiction. That is not a form you want to be reading for the first time while your app is down and your revenue is at zero. Read it once now, while it is abstract.
Where I could be wrong
The strongest counterargument is scale. Platforms hosting millions of apps genuinely cannot review notices manually, and the DMCA's safe harbor structure actively pushes them toward removing first and asking later, because that is what protects them from liability. Google is following the incentive the statute created. Blaming the automation alone lets the statute off the hook.
I also do not know what Tracer.AI's actual false positive rate is. Three publicized bad notices out of an unknown denominator could be a rounding error against a system that correctly removes thousands of genuinely infringing apps. That would not make these three filings acceptable, but it would make "AI takedown bots are broken" too strong a claim, and I do not have the data to support the stronger version.
What I do not think is arguable: a notice that names no infringing asset, filed twice against the same project, and a 46-day restoration against a 14-business-day statutory deadline. Those are facts about process, and the process is bad independent of the hit rate.
The honest take
Do not read this as a story about AI copyright enforcement being dumb, even though it is. Read it as a reminder that your distribution is a dependency, and it is the one dependency most solo operators never mirror. You would not run a production database with no backup. Most of us run a single store listing with no second channel, and it can be switched off by a company you have never heard of, on behalf of a company you do not compete with, for reasons nobody has to specify.
Go set up F-Droid, or a direct download, or a second store. It will take you an afternoon. Do it on an afternoon of your choosing.
Author
Lukas
@lukcombinator