We need to admit something uncomfortable: Open Source is under attack.
It’s not an attack by hackers (mostly). It’s an attack by helpful robots.
If you maintain a public repository, you know the feeling. You wake up to 10 new Pull Requests. The titles look promising: “Refactor utility functions,” “Optimize loop performance,” “Fix edge case in auth.”
You open the first one. It’s clean. The comments are polite. The code looks… plausible.
So you pull it. You run npm install. You run the tests.
And it explodes.
The “Green Square” Incentive
Why is this happening? Because we have combined two dangerous things:
- Gamification: Developers want “green squares” on their GitHub graph to look active.
- Zero-Cost Generation: AI agents can now generate a “contribution” in 30 seconds for $0.01.
The result? A firehose of low-effort, high-confidence garbage.
A recent study showed that AI-generated PRs are nearly 2x as likely to contain bugs as human ones. But that’s not the worst part. The worst part is that they are 10x harder to review.
The Hallucination of Competence
When a junior dev writes bad code, it looks bad. The formatting is off. The logic is messy. You spot it instantly.
When an AI writes bad code, it looks perfect. It follows the style guide. It uses the right variable names. It hallucinates a method that should exist, but doesn’t.
To find the bug, you can’t just read the diff. You have to run it.
And that is where the trap snaps shut.
My Laptop is Not Your Staging Server
Every time you git checkout an untrusted branch and run a build script, you are taking a risk.
Is this code just buggy? Or does postinstall.js scrape your SSH keys?
In 2025, “Reviewing PRs” has become “Auditing Supply Chain Security.” Maintainers are burning out because the cost of verification has skyrocketed, while the cost of contribution has hit zero.
We are acting as the human error-handlers for a million AI agents.
Zero-Trust Verification
We need a new rule for Open Source: Don’t touch the code until you see it run.
This is why I use PrevHQ for every public repo I maintain.
When a random contributor (or a bot) opens a PR, I don’t pull it. I wait for the PrevHQ bot to comment:
Preview Live: https://pr-402.prevhq.app
I click the link.
- Is the page blank? Close the PR.
- Is the console throwing errors? Close the PR.
- Does the “optimization” actually make it slower? Close the PR.
I do all of this from my browser. My terminal stays closed. My local environment stays clean.
Save Your Sanity
We cannot stop the flood of AI code. It’s here.
But we can build better dams.
Stop treating every PR like a gift you have to unwrap carefully. Treat it like a suspicious package. Put it in a blast chamber (a micro-VM), detonate it (run the build), and only look at what survives.
If you want to survive the Spam Apocalypse, you need to stop reviewing diffs and start reviewing reality.
Your time—and your SSH keys—are worth it.