They're calling it the first botnet designed for broadband equipment and routers, and that it is. But it's also the first of something else: Psyb0t is the first Linux botnet.
And even though it's running on hardware devices and even though it's running on Linux, and an obscure distribution of Linux at that, the basic mechanisms of it aren't that different from "conventional" botnets that run on Windows PCs. There's a lesson here.
Linux seems to be a great platform for these little embedded devices. It's small enough that it can fit in economical hardware, portable enough that you can put it on almost any processor and platform, and it's got great networking tools. This particular bot runs on Linux Mipsel devices ("Mipsel" refers to little-endian implementations on MIPS processors, generally, but not exclusively, on Linux). But it's not hard to see the same thing happening to any sufficiently large population of Internet-facing devices based on Linux or any other platform. I'm especially curious about DVRs now.
We often speak about how malware writers write for Windows because that's where the systems are and because that's where the development tools are, for malware and more generally. The same could be said now of Linux: The fact that a device runs Linux means it's easy to write binaries for it that do networking tasks, including hardening the bot and distributed denials of service.
How does Psyb0t work? The main vulnerability it seems to exploit is simply weak or nonexistent authentication. One involved device is the NetComm NB5 ADSL (asymmetric DSL) modem, earlier versions of which were administrable from the WAN side by default. In fact, some were administrable without any log-in at all. Of course updates were made, but when was the last time you applied an update to your ADSL router? I've seen vaguer reports of other vulnerabilities used.
According to DroneBL, the DNS (Domain Name System) blacklist service that found the botnet, Psyb0t appears to have been shut down just recently.
The bot will not persist if the router is power-cycled, but who does that on purpose? I also wouldn't discount the possibility that such a bot could be built to flash itself into an EPROM (erasable programmable ROM) or some other persistent memory, and then the device would probably be unsalvageable. Such an attack would be highly model-specific.