Comments on: Captcha Code

by: Gerardo Capiel

Wed, 05 Jan 2005 01:40:13 +0000

Are you making your source code available?

by: Mike

Sun, 27 Jun 2004 09:28:13 +0000

I wouldn’t call it a flaw. The reason your given the secret key, submitted hash, and correct hash, is because this is just a demo. I have changed the output of the script to make this a little more obvious. Thanks.

by: Oliver

Sat, 26 Jun 2004 14:53:46 +0000

I’ve found a flaw: as long as you give the calculated hash in the answer page, I can re-submit the answer with the hash you’ve given me (same input, but your hash). With this method I don’t even need to load any picture at all!

The first thing to do is remove the real hash on the answer page. I wrote a small script here to have a right answer every time.

by: Mike

Fri, 25 Jun 2004 20:35:44 +0000

The reason I intergrated the ’secret’ into the computation was to allow a site owner to significantly change the hashing function with a few keystrokes.

Assume for a second you do the 2^64 computations (thanks for doing the math Oliver) to find the secret. By changing a few characters in the secret, you again have to do 2^64 computations to find it.

I do appricate the feedback on my methods though. Because of this dicussion i’m moving the hashing secret into an external file so a cron job could be setup to generate a new secret every day/week/month/etc.

by: Oliver

Fri, 25 Jun 2004 00:06:14 +0000

I’m not an expert on this point, but using SHA-1 instead of MD5 wouldn’t strenghten the program. The cracker would just have to do 2^80 computations (for SHA-1) instead of 2^64 (for MD5): 2^64 is big enough. I even think it could have no effect at all if you just try to brute-force the program. By using SHA, you just garantee the number is unique.
Expanding the timestamp to seconds would be a good solution.
The best solution would be to show us the source code so we could analyze it!

by: Chas

Thu, 24 Jun 2004 19:59:25 +0000

Hmmm. A cracker wouldn’t necessarily be privy to this information, but knowing it would improve the chances of brute force cracking the S value. Once that was cracked, it would be trivial to reassemble or create a new TSH as you described. A good step would be using SHA, but I think expanding the timestamp to minutes (or even seconds) would solve the rest of the problem.

In an application like a whois lookup (NetSol uses a captcha for theirs) limiting to seconds would be much more effective than minutes. Overall, looks good. I don’t have any experience with character recognition, or I’d try that for you.