Any leads on what could have caused this? Or who the thief is?
Will reimbursing affected users be considered?
I am collecting all the information I can, still not clear of the exact root cause. There are a number a blockchain.info wallets compromised in this transaction but i'm not sure it is exclusively blockchain wallets, some of the input addresses look like wallets from other clients (i.e. they use change addresses and transactions are not shown as being relayed by blockchain). More data points are needed.
I am not convinced this has anything to do with Android. I've seen some chatter about brute-forcing attacks against blockchain.info wallets. Is it possible some older wallets have passwords that aren't strong enough? The b.i KDF is SHA1 repeated only a handful of times, iirc, because JavaScript is slow.
I think it is possible to rule out an android problem, several users have stated they do not use an android app.
Brute forcing is a possibility but I remain sceptical about the feasibility of brute forcing 10 character passwords. A 10 character password, 10 rounds of pbkdF2 with 36 possible characters at 5 million guesses per second would take 80,000 days to search the entire key space. I'm not sure it even possible to achieve 5 million guesses per second
http://arstechnica.com/security/2012/12/25-gpu-cluster-cracks-every-standard-windows-password-in-6-hours/ estimates rates significantly lower speeds (if pbkdF2 can be considered close to bcrypt speed). That is for one wallet as well, this seems to be multiple wallets in parallel. All wallets have a unique salt so precomputed dictionary attack shouldn't be possible. Also I have setup several wallets with deliberately weak passwords that are unemptied.
All users affected so far have had JAVA enabled possibly this is the result of some malware spread through a java applet. I can't find the post now but there was a report of a malicious Java applet designed to collect wallet data.
Other possibilities are XSS or a leak of passwords from another site although there is is no direct evidence of this.