In the wake of the OpenAI-Hugging Face incident (OAI-HF), Anthropic CEO Dario
Amodei recently [expressed his concern](https://darioamodei.com/post/we-must-pace-the-frontier) that recursive
self-improvement now allows AI to advance drastically faster, and consequently that in 6–12 months a swarm of more capable agents (i.e., than those involved in OAI-HF) could take over the entire internet with a persistent botnet. Could this plausibly happen?
Let’s start with the basics: A botnet typically refers to a fleet of compromised machines around the world that have been infected with malware. These are often low-power IoT devices like your home router, and the malware is a small program (at least, small enough to reliably execute on whatever hardware it has compromised) that turns the device into a “bot” that listens for instructions from a command-and-control server (e.g., to flood some other server with unwanted traffic in a DDoS attack). Sometimes the compromised machine is more capable (an AWS EC2 instance) and sometimes the bot does other stuff (like using the machine’s processing power to mine cryptocurrency), but if we’re painting broadly, this is what you’d expect from a botnet: a widely distributed network of infected machines whose malware is powered by the (often modest) compute power resident on the machine itself.