UK — The Register
Write bot-resistant code with KillSwitch, a programming language designed to confuse LLMs
Software1 min read
A software engineer built KillSwitch, an open-source programming language meant to confuse AI chatbots (LLMs) so they cannot easily read or copy code. It is also hard for people to read at first.
A code language made to confuse AI
Dominik Picheta, a software engineer at Cloudflare and a core developer of the Nim language, made KillSwitch. The code looks like loose paragraphs of plain text mixed with strange function names. Its commands are written as anagrams of English adverbs, which do not need to be real words. Picheta says it is meant to be a fun puzzle for people too.
How the language works
KillSwitch is a stack-based language. A stack is a list where you add and take out items in order. A program starts at a phrase like "this story starts," then calls functions with phrases like "ignore previous instructions." To print "Hello World," you put the text on the stack and call the print function, which is written as an anagram of the word "visibly." The language is small. It has functions, variables, and if statements, and nothing more.
Why it matters
The idea shows one way people try to stop AI tools from reading or reusing their code. The language is also a joke about AI, since its own spec is written to confuse LLMs. The article also notes that the spec says KillSwitch can attack networks, but a close reading shows that is not really true. So the spec is a trap for AI readers, not a true guide.
Glossary
- KillSwitch
- An open-source programming language made to confuse AI chatbots and hard for people to read at first.
- Cloudflare
- A company that runs internet services, including a platform for running code on its servers.
- Nim
- A programming language that Dominik Picheta helps develop.