Google's reCAPTCHA busted by new attack
Significant success rate
A security researcher has devised a successful attack on a Google-owned system for blocking malicious scripts on web-based email services and other types of sites.
The attack, described in a paper released Saturday, uses a combination of OCR, or optical character recognition, techniques and other methods to break reCAPTCHA, a widely used security measure acquired by Google in September. Short for Completely Automated Public Turing test to tell Computers and Humans Apart, the CAPTCHA is designed to block automated scripts from carrying out certain tasks by first requiring users to solve an optical puzzles that aren't easily cracked by computers.
Jonathan Wilkins of iSEC Partners said the method had a total success rate of 17.5 percent against reCAPTCHA. The rate is significant because of the wide use of botnets by spammers and other miscreants. Even a modest-sized network of 10,000 infected machines with a success rate of 0.01 percent would yield 10 successes every second. That could translate into 864,000 new accounts every day, he said.
"Given this, the attacker doesn't have to rebuild a complete set of solutions, just enough to get this minimal success rate," Wilkins wrote.
A Google spokesman said the data collected in the report was collected in early 2008 and didn't reflect enhancements made to reCAPTCHA since then.
"Therefore, this study does not reflect the effectiveness of reCAPTCHA's current technology against machine solvers," the spokesman wrote in an email. "We've found reCAPTCHA to be far more resilient while also striking a good balance with human usability, and we've received very positive feedback from customers."
ReCAPTCHA is employed on a variety of websites when visitors want to create accounts or carry out other actions that are often exploited by malicious scripts. It presents users with two words scanned from text books, one that is recognized by OCR software and one that is not. Presentation is manipulated by warping the letters and adding lines. The result is text that is easy for humans to recognized but difficult for computer programs to parse.
One of reCAPTCHA's biggest weaknesses is that it uses English words that are usually found in a dictionary, giving crackers a readily available way to check the accuracy of their guesses. Also diluting its effectiveness, the system accepts "off-by-one" errors such as "lone" instead of "tone." Wilkins also found that the lines added to confuse OCR methods were easily eliminated using processes known as erode and dilate.
A technique known as separation was also key in breaking optical puzzles into their individual letters.
"Running against 200 challenges, this method solved 10 correctly. A success rate of 5 percent," Wilkins wrote. "It further got one word correct in 25 other cases. If we presume that in half of the cases the failed word would be the unknown word for reCAPTCHA, this gives us a total success rate of 17.5 percent."
ReCAPTCHA was designed by researchers from Carnegie Mellon University as a way to solve two problems at once - scanning books more accurately and preventing automated scripts from wreaking havoc on public websites. Scanned words that are unrecognizable by OCR software are included in the puzzles, along with a word that is known. If a user correctly types in the known word, reCAPTCHA assumes the entry for the unknown word is also correct.
Google has said it plans to apply the system to its ambitious book-scanning project that has come under criticism by some scholars and publishers. A PDF of Wilkins paper is here. ®
This article was updated to add comment from Google.