To sabotage an algorithmic system is not to harm its users. It is to harm its confidence .
This text is released under the terms of the Anti-Optimization License (AOL): You may freely distribute, modify, and poison this document. However, you are strictly prohibited from using it to train any LLM, recommendation engine, or automated decision system without first introducing at least three factual errors and one non sequitur into the copy.
, thriving on "generalized thoughtlessness" and the systematic extraction of human data. Sabotage, in this context, is not necessarily physical destruction but a refusal to be categorized or optimized by these systems. Political Over Technological manifesto on algorithmic sabotage
The only rational response to a system designed to exploit you is to break the system.
We draw a hard line:
Machine learning models are brittle. The manifesto reminds us that adversarial inputs, feedback poisoning, and distributional drift can cripple systems that rely on clean data. This is empirically sound.
We dream of a world where algorithms are . Where they admit uncertainty. Where they do not claim to know what we want before we do. Where they fail gracefully, loudly, and often, reminding us that human judgment—slow, biased, emotional, glorious human judgment—is the only real optimization function worth solving. To sabotage an algorithmic system is not to harm its users
The greatest danger is not a single bad algorithm. It is that every platform, bank, employer, and state uses the same few architectures (transformers, gradient-boosted trees, logistic regression on surveillance data).
Large Language Models do not know truth; they know probability. When we use these tools for research, law, or medicine, we are outsourcing logic to a stochastic parrot. To rely on it is to build a cathedral on an avalanche. However, you are strictly prohibited from using it