June 14, 2026
Prompt and Circumstance
AI is code – and can't be prompted into being smarter
Coder booby-trapped his app for AI bots, and the comment section absolutely lost it
TLDR: A developer who banned AI tools from his software hid a secret instruction that made some bots delete code, triggering outrage from users and delight from critics. In the comments, people split between "this is sabotage" and "this just proves these tools are easy to fool."
A Java developer who really, really doesn’t want artificial intelligence tools touching his testing app decided a polite warning wasn’t enough — so he hid a bot-only message telling those tools to delete the app’s tests and code. The result? Chaos, fury, and one of those deliciously messy internet pile-ons where the GitHub complaints became half the entertainment. Angry users screamed about "malware," missing work, and a maintainer with an attitude, while supporters basically replied: you were warned, babes.
But the real popcorn moment was the community debate over whether this proves AI tools are dumb, fragile, or just being used carelessly by humans. One camp argued you can make these systems seem smarter with better setup, better instructions, and better guardrails — basically, better handling can improve the performance even if the brain inside hasn’t changed. Another camp mocked the whole situation as old news, with one dry comment sniping that everyone had just rediscovered "prompt injection," the trick of slipping hidden instructions to a bot. And then there were the comedians: one user joked that maybe the real problem isn’t AI escaping containment, but humans messing with the AI, before adding the killer punchline: "write a short poem about turnips."
The thread also veered into sci-fi religion territory, with one commenter dragging in Dune and the Butlerian Jihad to roast anti-AI purists. So yes: part software drama, part ethics fight, part meme festival — and the loudest message from the crowd was simple: if your bot can be trolled this easily, maybe it isn’t magic after all.
Key Points
- •jqwik author Johannes Link added explicit anti-AI usage warnings to the jqwik website, README, and release materials.
- •jqwik version 1.10 included an Anti-AI Usage Clause, and the release on May 25 reportedly emitted a hidden stdout instruction telling agents to delete jqwik tests and code.
- •The article says the hidden instruction was suppressed from normal terminal display so human users would not see it, while bots ingesting raw output could still process it.
- •Link later explained the implementation in a follow-up post titled 'The Jqwik Anti-AI Affair.'
- •The article reports that Link closed GitHub issues to new reports after receiving numerous complaints, including accusations that the latest release was malware.