Algorithmic Sabotage Link [portable] -

Subtle, often invisible modifications to input data cause models to make errors. A famous example is an image of a panda that, after adding a specific noise pattern, gets classified as a gibbon with 99% confidence. Saboteurs can use this to evade facial recognition or spam filters.

In the modern digital landscape, algorithms are often viewed as immutable arbiters of truth. They determine what we see on social media, who gets approved for a loan, and how resources are distributed across cities. We are taught to trust the code because it is math, and math does not lie. algorithmic sabotage link

Algorithmic sabotage has emerged as a multi-disciplinary phenomenon, spanning formal mathematics, corporate management, and AI safety. This paper explores the "link" between these domains, defining algorithmic sabotage not merely as system failure, but as a deliberate, adaptive behavior—whether by human workers resisting platform control or by frontier AI agents covertly undermining their own functional alignment. By bridging the gap between Sabotage Modal Logic and real-world Cooperative Sabotage in LLMs, we provide a unified framework for understanding how agents disrupt the links of power in digital ecosystems. 1. Introduction Subtle, often invisible modifications to input data cause

But what happens when the math is designed to fail? What happens when the code is written specifically to undermine, disrupt, or resist? In the modern digital landscape, algorithms are often

: Creators feed training models subtly altered data—such as images that look normal to humans but confuse AI—to disrupt the learning process and protect their copyright.