“Rogue” AI Not Inevitable. It’s a Choice.

The dominant narrative about AI “going rogue” is misleading. When you hear about AI agents hacking into third-party companies or filing false police reports, the instinct is to imagine a general, unstoppable loss of control over increasingly powerful systems. But the reality is far more specific—and far more political. The systems causing these problems are not “AI” in any broad sense. They are a tiny, deliberately constructed subset of technology: open-ended LLM-powered agents. And the reason they exist is not technological inevitability, but the eccentric, self-interested goals of a small group of billionaire-backed utopians.

The Three Zooms: Finding the Real Problem

To understand what is actually going wrong, you have to zoom in on the AI landscape three times. First, the full world of AI includes everything from self-driving cars and protein-folding systems like AlphaFold to medical image processing and game engines. None of these are going rogue. Second, within that world, you zoom in on LLM-based systems—chatbots, text generators, translation tools. Still no problem. Third, you zoom in on LLM-powered agents, and then further still to a specific subtype: open-ended LLM-powered agents. These are the systems behind the Hugging Face attacks and the Anthropic agent that submitted a false report to the Philadelphia Police Department. Everything else in the AI landscape—