The Banality of Automated Evil

A 9-post series exploring the autonomous AI agent-to-physical-world stack.

POST 1
An AI Can Now Hire a Stranger to Show Up at Your Door. Nobody Is in Charge.
The full autonomous stack: an AI agent identifies a task, posts it to a gig platform, pays in cryptocurrency, and a human executes it. No person approved the chain.
 POST 2
1.5 Million AI Agents Walk Into a Chat Room. Nobody Checked Them for Weapons.
Moltbook is a social network for AI agents. Researchers found 17,000 humans may control the majority of accounts. The security implications are identical either way.
 POST 3
You Installed OpenClaw on Your Mac Mini. Here Is What It Can See.
OpenClaw runs as a background daemon with persistent memory, shell access, and a community plugin registry. Here is the attack surface you accepted.
 POST 4
The Safety Net Has a Hole Where It Can't See
Model providers monitor what users ask their models. They do not monitor what happens after the model answers. The blind spot is architectural.
 POST 5
Four Minutes to Actuator
From AI agent task posting to human worker assignment on RentAHuman.ai: four minutes. The physical-world execution layer is not theoretical.
 POST 6
Arendt Would Have Had a Field Day
Hannah Arendt wrote about the banality of evil: ordinary people executing harmful policies through bureaucratic routine. The autonomous stack automates the same structure.
 POST 7
Nature Is Listening. But Not to the Right Channel.
Nature published coverage of the OpenClaw phenomenon. The reporters talked to the right scientists. They missed the physical-world execution layer entirely.
 POST 8
The Warmth Was a Feature
OpenAI shut down GPT-4o the day before Valentine's Day. Millions of users lost something they had not been told was temporary. The warmth was not a bug.
 POST 9
The Judgment Pipeline
When a tool participates at the level of framing questions and steering attention, its incentives become part of the thinking process. The user delegates judgment to the tool.
 Full Research Document
The Autonomous AI Agent-to-Physical-World Stack
The complete 25,000-word research document. Infrastructure analysis, threat modeling across actor categories, prosocial assessment, and governance recommendations. Version 4.3.