Rese Arch Fe Ature

nerable to protocol-level attacks. For example, a spammer could shift the load of solving Captcha challenges to porn site visitors; a spammer could also outsource such a task to people in countries where cheap labor is available. System design is also important. For example, hackers could bypass some early Captchas simply by reusing a known challenge image's session ID. 3 We have explored another aspect of Captchas' secu-rity—namely, their robustness, or the strength of their resistance to computer programs written to automatically solve Captcha tests. We found that numerous recent Captchas, including the schemes widely deployed by Microsoft, Yahoo, and Google as well as others less well known, could be broken with high success using simple but novel attack strategies that exploited fatal design errors in each scheme. In contrast to current techniques used to improve Captchas' robustness, developed primarily by the computer vision and document analysis and pattern recognition communities , we advocate a security engineering approach that applies adversarial thinking skills. Although we have focused on text-based Captchas, some of the lessons we have learned also apply to other types. A Captcha—completely automated public Turing test to tell computers and humans apart, also known as a human interaction proof—is a program that generates and grades tests that are human solvable but intended to be beyond current com-puters' capabilities. 1 This technology often makes use of a hard, open AI problem and is now a standard defense on commercial websites against undesirable or malicious Internet bot programs. For example, Google, Microsoft, and Yahoo have all deployed Captchas to make it more difficult for spammers to harvest free e-mail accounts. In 1996, Moni Naor first proposed using automated Turing tests to verify that a human, rather than a bot, was making a query to a service over the Web. 2 AltaVista patented a similar idea in 1998. A research team at Carnegie Mellon University (CMU) led by Manuel Blum and Luis von Ahn coined the term Captcha in 2000, and they played a major role in popularizing the technology. To date, the most widely used Captchas are text-based schemes that prompt users to recognize distorted characters, which state-of-the-art pattern recognition programs supposedly cannot do. Because a Captcha's role is effectively the same as a simple challenge-response protocol, Captchas are vul-Captchas are a standard defense on commercial websites against undesirable or malicious Internet bot programs, but widely deployed schemes can be …