Case Studies

Luanti Yanked From Google Play by an AI Company's Bogus DMCA

A veteran open-source voxel engine vanished from Google Play after a takedown notice signed by "Tracer AI." The Luanti team's reverse investigation found the sender held no copyright at all, exposing how little vetting platforms actually do.

Font size

On August 27, Luanti (formerly Minetest) disappeared from Google Play.

The removal wasn't voluntary. A DMCA takedown notice pulled it down, signed by Tracer AI, claiming the project's code and assets infringed on the company's model training data. Google followed the standard playbook: take it down first, notify the developer later. The whole thing took under 48 hours.

The Luanti team didn't wait around. Over the next few days they ran their own reverse investigation and concluded the notice was completely baseless.

The story made the front page of Hacker News, sitting at 132 points with 39 comments (link at the bottom). But the more important story isn't the HN thread. It's the systemic gap the incident exposes.

What Happened: One Email, One Vanished App

Luanti is a 15-year-old open-source voxel game engine, community-driven and licensed under LGPLv2.1+. It had been on Google Play for years and was something of a benchmark for "open-source game demos."

On August 27, a notice appeared in the Google Play developer console: a DMCA takedown claiming that Luanti's code or assets infringed on model training data. The claimant was listed as Tracer AI.

Google's standard procedure runs like this:

  1. Receive notice → remove immediately
  2. Tell the developer "someone filed a claim against you"
  3. Developer can submit a counter-notice to appeal
  4. Platform does not verify whether the notice is legitimate

That process is baked into Section 512 of the DMCA, commonly called the "safe harbor" provision. As long as the platform follows the procedure, it avoids secondary liability. The trade-off: all the verification work falls on the accused.

The Luanti team's official blog documents the incident in detail: blog.luanti.org/2026/08/27/luanti-dmca-tracer-ai/. By the time the blog post went live, Luanti's app page already showed "Not available."

How Luanti Proved Its Innocence

The team didn't wait for the appeal window to open. They did the detective work themselves.

The notice itself contained almost nothing: just a name, "Tracer AI," and an email address. From that thin thread, they ran several checks.

1. Reverse-lookup on the sender's identity. "Tracer AI" turned up no matching company. A WHOIS lookup on the domain showed it had been registered very recently, and the email used a disposable email service. Hardly the profile of a legitimate business; more like a throwaway alias.

2. Verifying the supposed infringing content. The notice vaguely claimed "model training data was infringed." Luanti is an LGPL game engine. The code is C++ and Lua; the assets are user-created models and textures. Where exactly the training data came from was never explained.

The team went through the repo's full history, commit records, contributor agreements, and issue tracker. Nothing traced back to "Tracer AI."

3. Checking copyright registrations. A search of the U.S. Copyright Office's public database (cocatalog.loc.gov) turned up zero records under Tracer AI. An "AI company" without a single software copyright registration, filing DMCA notices.

4. Filing the counter-notice. Eventually the team submitted a formal DMCA counter-notice. Google restored the listing seven days later. During those seven days, the app was completely unreachable. For a free, open-source project, that's effectively a blackout.

Luanti published the full investigation as a public report that anyone can reproduce. That kind of transparency is pretty on-brand for the project.

DMCA's Design Flaw, Amplified by the AI Era

Section 512 of the DMCA was written in 1998. Back then, the internet mostly meant websites and FTP servers, and the piracy problem was MP3s and movies. The legislative logic went like this:

Too many platforms, too much piracy, too few rights-holders to chase every case. Let platforms "remove first, be shielded from liability later." Maximum efficiency.

The trade-off: the filer doesn't need to prove they actually hold the copyright. DMCA counter-notices contain a line that reads, "I have a good faith belief that the material was removed as a result of a mistake or misidentification." In practice, that means all you have to do is sign a statement saying "I believe this is piracy" and the platform takes it down.

In the AI era, this has become a ticking bomb.

Historically, DMCA was mostly used by music labels and film studios, entities with clear, traceable copyright chains. AI companies don't work that way:

  • Training data is a black box; you can't pinpoint "which line of code" was infringed
  • Statements like "my model was trained on your code" are nearly impossible to disprove
  • The number of AI companies has exploded, making malicious DMCA filings extremely cheap: one email, one day of effort

In the Luanti case, Tracer AI is obviously a shell. But even if it were a real AI company, the claim "my training data was infringed by your code" is technically almost impossible to rebut quickly.

Worse, the filer can stay anonymous, fabricate an identity, and skip the details. Takedowns are instant; rebuttals are delayed. During the gap, the app simply doesn't exist.

Google Play's Responsibility Vacuum

Luanti isn't the first casualty, and it won't be the last.

Google's current handling looks like this:

  • No identity verification on the filer: if the email sends, it counts
  • No detailed notice to the developer: just "someone claimed something"
  • No SLA on counter-notice review: how long it takes depends on the platform's mood
  • No history-based filtering: the same filer can repeatedly target apps without getting flagged

This setup is maximally convenient for Google. It stays legally shielded, operationally lazy, and can always point to "we followed the law" when pressed.

For open-source projects, the consequences are fatal. Distribution channels are heavily concentrated: Google Play, Apple App Store, Steam, itch.io. If any one of them gets jammed, the project grinds to a halt.

No license in the world, whether MIT, Apache, or LGPL, saves you here. The code itself isn't banned. Nobody can download it.

This isn't about open-source licensing. It's about platform economics.

Editorial: Open Source Stays Naked Under Platform Economics

The story here isn't "AI companies abusing copyright." It's that open-source projects are permanently exposed under platform economics.

DMCA is a 1998 law built on the assumption that rights-holders are clearly identifiable entities. In the AI era, that assumption collapses. A rights-holder can be anyone who types a single prompt. An infringement claim can be any vague "my model was trained on your stuff."

My contrarian prediction: in the next 12 months, we'll see more malicious DMCA filings made in the name of AI. The reasoning is straightforward:

  1. Cost is low: one email
  2. Anonymity is easy: WHOIS won't help
  3. Platforms have no incentive to change
  4. Legal precedent hasn't caught up

Google almost certainly won't change anything voluntarily. Its safe harbor protections are intact, and its motivation to act is zero. Unless a major class-action lawsuit or regulatory fine forces the issue, the status quo holds.

What's different this time: past DMCA abuse was mostly "overreach" by legitimate rights-holders making broad claims. Now it's "fabrication": claims with no underlying rights at all. In Luanti's case, the filer didn't even have a software copyright registration. A decade ago, someone pulling this stunt would at least have forged a registration number. The "AI" label is too convenient: it sounds intimidating, stays vague, and offers no concrete allegation a platform can evaluate.

What the open-source community can do is limited:

  • Back up distribution channels in advance: self-hosted GitHub releases, self-hosted APKs, mirrors on APKPure, etc.
  • Add mirror links to the project homepage
  • If taken down, publish a public statement immediately so public pressure pushes the platform to act

All of this is defense, not a solution. The actual solutions are two: either legislation that forces platforms to perform minimum identity verification, or the open-source ecosystem walking away from centralized distribution platforms entirely.

The first won't happen in five years. The second won't happen in five years either.

So for the next five years, open-source projects are on their own.


References

Source: https://blog.luanti.org/2026/08/27/luanti-dmca-tracer-ai/