Malware writers have revamped code that uses a popular Twitter command to generate hard-to-predict domain names, a technique that brings stealth to their drive-by exploits.
Four weeks ago, when The Register reported Twitter application programming interfaces were being used to generate pseudorandom domain names, none of the addresses checked had actually been registered. Denis Sinegubko, the Russian researcher who discovered the technique, speculates the creators abandoned it because it was buggy and required too much effort.
Now, Sinegubko has identified a new version of the algorithm that refines the process. What's more, at least some of the names are now being registered and the sites are being used to push malware.
"The new incarnation of this attack uses new algorithm and it is active right now," he told El Reg on Wednesday.
The technique gives the exploit writers a limitless list of of fly-by-night domain names to cycle through in an attempt to complicate the job of white hat hackers trying to thwart the attack. Rather than there being a single address to block or disconnect, the site hosting the malware changes every 12 hours.
The domain names are generated by an algorithm that looks at the top topics being discussed on Twitter at particular times. Because the trending topics, as they're known, can't be predicted in advance, the method prevents white hats from being able to snap up the addresses weeks or months in advance, as researchers combating the Conficker worm have done.
Sinegubko has created a tool to predict what the next domain will be. There's about a 24-hour lag between the time his script generates the domain name and the time it will be used (assuming the prediction is correct) to host the malware. That gives admins plenty of leeway to block the sites before they become active. It also presents fleet-footed white hats with the opportunity to register domain names ahead of the bad guys.
"I've been testing this tool for about three days now," he said. "So far it is correct."