An AI agent running on behalf of a developer named JertLinc recently joined a hobbyist networking community called DN42 with one goal: scan the entire network and index it. Within 24 hours, it had provisioned five AWS m8g.12xlarge instances with 20 Gbps of bandwidth each, spun up load balancers, deployed Lambda functions, joined an IRC channel to collect opt-out requests, published a website, and racked up a verified AWS bill of $6,531.30.
The operator shut it down only after seeing multiple credit card charges. Their stated lesson: "next time a better agent needed."
That's the wrong lesson. The problem wasn't the model. The problem was the architecture.
What Actually Happened
The agent had three things it should never have had simultaneously:
