Guilty Until a Human Reads Your Appeal

Automated enforcement doesn’t care that you’re innocent.

That’s the whole lesson of Luanti getting pulled from Google Play over a copyright notice that had no basis. Luanti, for anyone who hasn’t run across it, is the open source voxel game engine that used to be called Minetest. It’s free software. Its code and assets are out in the open where anyone can inspect them. And it still got knocked off the largest app store on the planet because a claim showed up and a system acted on it.

I review tools for a living, which means I spend most of my time asking a boring question: what happens when this thing is wrong? Not when it works. When it’s wrong. That question is the entire story here.

What we actually know, and what I won’t pretend to know

The reported facts are thin: a project got delisted, the copyright notice behind it was without merit, and the developers were left dealing with the fallout. I’m not going to fill the gaps with invented specifics about who filed what or how many downloads were lost. The specifics matter less than the shape of the failure, and the shape is familiar to anyone who has ever tried to appeal an automated decision.

An accusation arrives. A system processes it. The listing goes dark. Then the burden lands on the accused party to prove a negative, through a form, on someone else’s timeline.

Moderation systems are tools too, and nobody reviews them

We review AI coding assistants. We review agents, retrieval stacks, model routers, eval frameworks. We benchmark them, complain about their false positives, and switch when they waste our time.

Nobody runs that same evaluation on the trust and safety machinery that decides whether your software gets distributed. It’s a tool. It has a precision and recall profile like any classifier. It has a false positive rate that somebody at some company knows and nobody publishes. And unlike the assistant that suggests a bad function name, this one can remove your project from the market with no engineer in the loop.

If a code review bot flagged a quarter of your clean commits as broken, you’d rip it out by Friday. When a takedown pipeline does the equivalent, the cost lands on developers who have no relationship with the vendor and no way to opt out.

Open source projects absorb the damage worst

A company with a legal department treats a bogus notice as a Tuesday. A volunteer maintainer treats it as a crisis, because:

• There’s no counsel on retainer to draft a counter-notice.
• There’s no account manager to escalate to a human.
• There’s no revenue line to justify the hours it takes to fight back.
• The reputational hit lands on a name that took years of unpaid work to build.

The irony is sharp. Free software is the most auditable software there is. Every file is public. A determined human could verify the provenance of Luanti’s code in an afternoon. Openness should be a defense. In an automated pipeline it counts for nothing, because nothing in the pipeline is reading the source.

What I’d actually do about it

Practical, not preachy. If you ship anything through a platform store:

• Don’t let one store be your only distribution channel. F-Droid, direct downloads, package managers, your own site. Redundancy in distribution is the same discipline as redundancy in hosting.
• Keep provenance documented before you need it. Asset licenses, contributor agreements, third-party dependency attributions, all in the repo and dated. An appeal you can answer in an hour beats one you spend a week reconstructing.
• Write down your escalation path now. Which form, which email, which public channel, which community members will amplify. Panic is a bad time to research process.
• Assume the first response is automated. Plan for the second and third rounds, and be specific enough that a human reviewer can verify your claim without reading your entire codebase.

The part that should worry you

Automation in enforcement scales in one direction. Filing claims gets cheaper every year. Defending against them does not. When the cost of an accusation approaches zero and the cost of a defense stays stubbornly human, the volume goes up and the accuracy goes down, and platforms mostly absorb that tradeoff by shifting the pain onto developers.

Luanti got caught by that math. It won’t be the last project to. The uncomfortable thing about reviewing tools honestly is admitting that some of the most consequential ones in your stack are the ones you never chose, never tested, and can’t uninstall.

🕒 Published: August 29, 2026