KillSwitch: adversarial esoteric language benchmarked against LLMs
Overview
dom96 releases KillSwitch, an open-source esoteric stack-based language designed to confuse LLMs and coding agents.
It includes hidden lies in the spec and letter-counting rules in identifiers. In the author's benchmark, Claude Opus 5 scores highest at 66.8, and the best open-weight model is MiMo-V2.6-Pro. The author says the tasks show model performance but do not clearly prove that the models understand the language.
Written by AI from the articles below · updated Oct 19, 3:26 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- Hacker News · AI (150+ points)BlogKillSwitch: a programming language built to confuse LLMs and coding agents
AIThe author releases KillSwitch, an open-source esoteric stack-based language designed to be adversarial to LLMs and coding agents, with hidden lies in the spec and letter-counting rules in identifiers. Claude Opus 5 scores best on the author's benchmark with 66.8, and the best open-weight model is MiMo-V2.6-Pro. The author says the simple tasks show performance but not clear proof that models understand the language.
Heat Trend
Not enough continuous observations to show a trend yet.