Korelogic Logo
"Crack Me If You Can" - DEFCON 2012
  Team Hashcat has won the contest!  
Back to [Teams] [Top]

Team hashcat


Active Members 19
Nicks atom, d3ad0ne, radix, hbpanda, superjames, dakykilla, Xanadrel, K9, blaz, pure_hate, blandyuk, Rolf, kartan, RuraPenthe, |m|, chancas, ToXiC, iphelix, unix-ninja
Software oclHashcat suite, disthc, custom scripts, Passcovery Suite, John the Ripper
Hardware 83+ CPU cores (+ some hyperthreads), 54 GPUs


The initial feeling this year was that the contest had become overly complicated. The KoreLogic team introduced several new rules which seemed designed to handicap the larger teams, while we definitely appreciate the idea of getting more people involved in password cracking, as a large team, we felt rules such as those to be biased. Additionally, this year we did not feel the passwords and hash types were an accurate reflection of real world situations. Most of us as security professionals, have tools and mindsets that are crafted to attack hashlists which we find on penetration tests or publicly released list. For example the average length of the passwords this year was 21 characters (See: CrackMeIfYouCan Twitter Status) which is contrary to all data from recent breaches. Statistically the average password length is eight (8) characters. There was also no underlying theme to the passwords as there were in previous years, this goes against what is seen in most password breaches. After some time it became obvious that passwords where just sentences from books which unrealistic. This year's contest also seemed to be keeping with the theme of favoring John the Ripper as all algorithms were supported by this tool, which can't be said for Hashcat and PasswordsPro. We would like to extend our thanks to KoreLogic for taking the time and effort out of their otherwise busy work schedules to facilitate this contest. While we did have a few complaints this year, we do realize the substantial amount of effort that goes into creating a contest of this magnitude. We imagine that making this appealing to a wide range of individuals can be challenging and again we appreciate KoreLogic taking the time to put this together. In the past KoreLogic has always been receptive to contest feedback and we can only assume the same will be true for this year. As always we had a good time competing and the level of effort all of the teams put into the contest was amazing. The Hashcat team loves competition and is looking forward to next year's contest.

Contest Breakdown

Initially we started the contest in a similar fashion to previous years. We set out to crack as many passwords as we could by applying wordlists with common mutations. After the initial batch of easy hashes was cracked, the team regrouped and analyzed the current plains to determine a pattern or common theme. In previous competitions, KoreLogic had generated passwords by mutating plains in common ways seen in industry breaches. This year however plains where mostly generated by using phrases from a wide range of books, novels, and screenplays. This put us at a disadvantage, as we weren't able to employ techniques such as Markov attacks, or rule mutations commonly used for attacking passwords. This also left some of us in a moral dilemma as we would essentially have to pirate authors works to obtain phrases and words that we needed. Disappointed, the team set off to download parts of novels, and text from the Internet and reconstruct phrases that would be used as flat dictionaries. We quickly identified clues which books were used (Lord of the Rings Trilogy, Pride and Prejudice, and others). Atom created some scripts which helped us to create the pass phrase dictionaries. They can be downloaded here:



Several other lists the Hashcat team used can also be downloaded:






Some rules:



Thanks to superjames, hash types were organized and loaded into a homegrown management system dubbed "List Condense", coded specifically for this competition. Recovered plains were uploaded to the website where they went through a quick verification process, and if passed, removed from the list and added to a dictionary. Recovered hashes, remaining hashes, and plaintext dictionaries were all updated in real-time. Members were able to download left hashes easily to minimize work being duplicated.

Rolf bossed the challenges right away, putting us up a quick six (6) in the first hour of the CMIYC contest. Custom scripts, no matter how good, will never do that. The software he used to crack 6 of the challenges is called Passcovery Suite, coded by a brilliant russian known as IvanG. It's very optimized and has built-in attacks, which brought us success, along with a private wordlist. This was then knocked back down to 4 due to a PGP key submission error.

Once everything was sorted we were sitting happily with four (4) first cracks, and most of the hashes in List Condense application ready to be worked. The team quickly rose to the top banging out the easy stuff and identifying sources to use as lists. Unix-Ninja had also been working on a server/client solution for distributed cracking (disthc) which allowed us to queue sessions, by distributing the work load, and dictionaries via the Internet.

It would seem KoreLogic tried to even the odds between teams by introducing an obscure algorithm sun-md5, in addition to weighing heavily on bcrypt and sha512crypt. We had an ace up our sleeve though, as atom had added both bcrypt and sha512crypt to a non-public version of oclHashcat-plus prior to the start of the competition. Radix and d3ad0ne used their larger GPU powered machines to focus efforts on these hash types. This worked amazingly well until we started seeing passes greater than 15 characters at which point we were forced to fall back to CPU's due to a limitation in oclHashcat-plus of fifteen (15) characters. The other high scoring hash type was sun-md5 which many of us had never even seen. This put us in a bit of a bind being worth 3000 points each, thankfully iphelix threw together a cracker in python and away we went slowly applying our word lists. For the most part, we didn't even bother with these as it was apparent that bcrypt, sha512crypt, md5crypt were going to be what won the most base points. Iphelix generated some rules using rulegen.py which will ship with the next PACK release. The new tool uses the Levenshtein-distance algorithm to automatically generate rules or better said, to regenerate rules. This way we thought if we only know the source (Dictionary) that KL used and that was later mutated using rules like it was in the last years we can regenerate the rules KL used and regenerate the full plain list.

Deciding to employ a bit of PsyOps, we held back some of our high value hashes to see if someone would do a big submit to take the lead. Inside Pro was first to bite and join us. All three (3) teams played cat and mouse with uploads for a while then we dropped the bomb. This gave us a comfortable lead, but all good things come to an end. The John-Users and Inside Pro submissions started adding up and we were hitting walls trying to figure out how phrases had been cut out of the books. Down to the last thirty (30) minutes got pretty hairy. Inside Pro appeared to have hit a wall and the JTR team was trickling plains in slowly overtaking us in raw points. Down to the last ten (10) minutes it was just a question of JTR having enough cracks to take the bonus points. We pushed our last submissions and powered down to see how KoreLogic scored it.

atom AMD FX8120 AMD hd6990 Ubuntu 12.04 x86_64
blandyuk AMD FX8120 AMD hd5850 Win 7 x64
blaz i7 2x AMD 7970 Win 7 x64
chancas i7 2x AMD 7970 Win 7 x64
d3ad0ne i7 980x, 2x Xeon x5650 4x ATI 6970, 8x ATI 7970 Ubuntu 12.04 x86_64
hbpanda AMD FX6100 AMD 5870 Ubuntu 12.04 LTS
iphelix unknown unknown unknown
k9 i5 2500K ATI 6970 Win7 x64
radix 2x Xeon 5645, i7 3770K 7x Nvidia 580 GTX, 2x AMD 7970 Ubuntu 12.04 x86_64, Windows 7 x64
Rolf FX-8150 2x NVidia 480 GTX unknown
superjames i7 4x ATI 6990 GPU donated by Ranvik Debian x64
T0XlC Xeon E5502 ATI 5870 Win7 x64
xanadrel i7 3820, i3-2130 2x ATI 7970 Win7 x64, Ubuntu 12.04 x86_64
|m| unknown unknown unknown
dakykilla 2x AWS XL, unknown 2x ATI 6990 Cent OS 64 bit
pure_hate unknown 8x Nvidia 580 GTX Ubuntu 11.10
RuraPenthe unknown unknown unknown
dropdead AMD X2, i7 920 AMD 5970, Nvidia 570 GTX Ubuntu 12.04 x86_64


Please contact us if you would like more information about our services, tools, or careers with us.
Privacy Policy : Copyright 2012. KoreLogic Security. All rights reserved