Sabotage” - “algorithmic

Today, quant funds spend millions on "adversarial robustness"—training their AIs to ignore sabotage. But it is an arms race. For every defensive algorithm, there is a saboteur building a slightly more clever liar. Let’s get pragmatic. You are a mid-level manager at an Amazon warehouse. The algorithmic management system (the "Hourly Fulfillment Index") has just flagged you for "idle time" because you took a 4-minute bathroom break. Your productivity score drops. You are one strike from termination.

But corporations don't want paranoid algorithms. They want confident ones. And confidence is exactly what saboteurs exploit. We will not eliminate algorithmic sabotage. We will learn to live with it, just as we live with bacteria. “algorithmic sabotage”

Consider the gig economy. Uber drivers have long engaged in "algorithmic jiu-jitsu"—accepting rides and then driving slowly, or collectively logging off during surge pricing to force a higher multiplier. These are acts of labor resistance, but they are also sabotage. They break Uber’s promise of "reliable ETAs." Let’s get pragmatic

Because

At first, leadership blamed a glitch. But after a forensic audit, the truth emerged: a disgruntled data scientist had poisoned the training set. He had inserted a few thousand "ghost trips" into the historical data. The algorithm didn't know it was being lied to. It simply learned that circling a block was an efficient way to kill time before a phantom pickup. Your productivity score drops

The Ghost in the Machine: How We Are Entering the Age of Algorithmic Sabotage

A system that cannot be questioned—a system that treats every input as truth—invites sabotage. By removing human discretion, we force humans to communicate with the system only through actions. And when the only language left is action, the action becomes violent (or deceptive).