UK researchers have devised a novel and inexpensive way of cracking Microsoft's Windows Live Captchas with a success rate of more than 60 percent, a finding that further exposes weaknesses in a key measure designed to keep miscreants from infiltrating free online services.
In a paper (PDF) published Monday, Jeff Yan and Ahmad Salah El Ahmad, of the School of Computing Science at Newcastle University, lay out a way to crack the Captcha used to protect Windows Live Hotmail. Using custom-written software, a standard desktop computer was able to correctly read the characters more than 60 percent of the time. Microsoft designed it with the goal that automatic scripts should not be more successful than 0.01 percent of the time.
The paper follows research released last week by Websense Security Labs that tracked automated bots in the wild that were creating random Hotmail accounts after cracking the Captcha.
While attacks on Captchas deployed by Microsoft, Google and Yahoo are nothing new, the latest research appears to show new strides in the breaking of such protections. Short for "completely automated public Turing test," a way of distinguishing between computers and humans, most Captchas require end users to identify the letters depicted in a highly distorted image designed to be unreadable by computer scanners.
In many of the previous attacks - for instance, one against Hotmail that was observed by Websense in February - it was unclear if there was cheap human labor that was reading the Captcha images, and in any event, the scripts were successful no more than 35 percent of the time. Websense observed similar attacks on Gmail that succeed only about 20 percent of the time. A Google software engineer contends the attacks are being carried out in Russian sweatshops.
In January, researchers reported successfully cracking Yahoo's Captcha. According to Yan, Yahoo updated its Captcha last month to make it more resistant to attack.
The Newcastle researchers took a decidedly different approach. They figured out a way to isolate each of the eight characters that make up a Hotmail Captcha image. Defeating Microsoft's so-called segmentation-resistant technology was a major accomplishment. It blends the characters together in an attempt to thwart optical character recognition. Once they were able to segment the image - usually in about 80 milliseconds using a PC with a Core 2 and 2 GB of random access memory - the machine could easily read the individual characters.
"For the first time, we have shown that although Microsoft's MSN Captcha intentionally bases its robustness on segmentation resistance, it is vulnerable to a simple, low-cost segmentation attack," the researchers wrote. "Therefore, our work shows that the MSN scheme provides only a false sense of security."
The latest attack observed by Websense seems to make similar strides. Scripts obtained by company researchers were able to successfully respond a Captcha challenge in about six seconds, leading them to deduce that the recognition is happening automatically, rather than relying on a human being.
"We don't know exactly how the MSN one is done," said Dan Hubbard, vice president of security research at Websense. "They're getting quite a bit more automated."
What we do know is that up until now Captchas have been an important defense in the never-ending battle against spam. Already, anti-spam services have begun throttling messages sent from Gmail and Yahoo in response to the growing abuse of such services. If you think the spam epidemic is bad now, just wait until these new cracks go mainstream. ®