Over the last decade, it has become well-established that a captcha's ability to withstand automated solving lies in the difficulty of segmenting the image into individual characters. The standard approach to solving captchas automatically has been a sequential process wherein a segmentation algorithm splits the image into segments that contain individual characters, followed by a character recognition step that uses machine learning. While this approach has been effective against particular captcha schemes, its generality is limited by the segmentation step, which is hand-crafted to defeat the distortion at hand. No general algorithm is known for the character collapsing anti-segmentation technique used by most prominent real world captcha schemes.

This paper introduces a novel approach to solving captchas in a single step that uses machine learning to attack the segmentation and the recognition problems simultaneously. Performing both operations jointly allows our algorithm to exploit information and context that is not available when they are done sequentially. At the same time, it removes the need for any hand-crafted component, making our approach generalize to new captcha schemes where the previous approach can not. We were able to solve all the real world captcha schemes we evaluated accurately enough to consider the scheme insecure in practice, including Yahoo (5.33%) and ReCaptcha (33.34%), without any adjustments to the algorithm or its parameters. Our success against the Baidu (38.68%) and CNN (51.09%) schemes that use occluding lines as well as character collapsing leads us to believe that our approach is able to defeat occluding lines in an equally general manner. The effectiveness and universality of our results suggests that combining segmentation and recognition is the next evolution of catpcha solving, and that it supersedes the sequential approach used in earlier works. More generally, our approach raises questions about how to develop sufficiently secure captchas in the future.

/pdf/the-end-is-nigh-generic-solving-of-text-based-captchas-3w55ws65ax.pdf

The end is nigh: generic solving of text-based CAPTCHAs

Since their inception, captchas have been widely used for preventing fraudsters from performing illicit actions. Nevertheless, economic incentives have resulted in an arms race, where fraudsters develop automated solvers and, in turn, captcha services tweak their design to break the solvers. Recent work, however, presented a generic attack that can be applied to any text-based captcha scheme. Fittingly, Google recently unveiled the latest version of reCaptcha. The goal of their new system is twofold, to minimize the effort for legitimate users, while requiring tasks that are more challenging to computers than text recognition. ReCaptcha is driven by an "advanced risk analysis system" that evaluates requests and selects the difficulty of the captcha that will be returned. Users may be required to click in a checkbox, or solve a challenge by identifying images with similar content. In this paper, we conduct a comprehensive study of reCaptcha, and explore how the risk analysis process is influenced by each aspect of the request. Through extensive experimentation, we identify flaws that allow adversaries to effortlessly influence the risk analysis, bypass restrictions, and deploy large-scale attacks. Subsequently, we design a novel low-cost attack that leverages deep learning technologies for the semantic annotation of images. Our system is extremely effective, automatically solving 70.78% of the image reCaptcha challenges, while requiring only 19 seconds per challenge. We also apply our attack to the Facebook image captcha and achieve an accuracy of 83.5%. Based on our experimental findings, we propose a series of safeguards and modifications for impacting the scalability and accuracy of our attacks. Overall, while our study focuses on reCaptcha, our findings have wide implications, as the semantic information conveyed via images is increasingly within the realm of automated reasoning, the future of captchas relies on the exploration of novel directions.

/pdf/i-am-robot-deep-learning-to-break-semantic-image-captchas-tms7xcg6qx.pdf

I am Robot: (Deep) Learning to Break Semantic Image CAPTCHAs

Text-based Captchas have been widely deployed across the Internet to defend against undesirable or malicious bot programs. Many attacks have been proposed; these fine prior art advanced the scientific understanding of Captcha robustness, but most of them have a limited applicability. In this paper, we report a simple, low-cost but powerful attack that effectively breaks a wide range of text Captchas with distinct design features, including those deployed by Google, Microsoft, Yahoo!, Amazon and other Internet giants. For all the schemes, our attack achieved a success rate ranging from 5% to 77%, and achieved an average speed of solving a puzzle in less than 15 seconds on a standard desktop computer (with a 3.3GHz Intel Core i3 CPU and 2 GB RAM). This is to date the simplest generic attack on text Captchas. Our attack is based on Log-Gabor filters; a famed application of Gabor filters in computer security is John Daugman’s iris recognition algorithm. Our work is the first to apply Gabor filters for breaking Captchas.

/pdf/a-simple-generic-attack-on-text-captchas-25fykhmypo.pdf

A simple generic attack on text captchas

A method and system for optimizing segregation between human-operated clients and machine-operated clients accessing computing resources are provided. The method comprises receiving, from a client, an authentication request, wherein the authentication request is received in response to a redirect request sent from a remote server to the client; dynamically selecting at least one authentication challenge from a plurality of different authentication challenges; sending the at least one generated authentication challenge to the client; determining whether a notification call is received from the client during a predefined time interval; and upon receiving the notification call during the predefined time interval, confirming that the client passes the authentication challenge, wherein a client that passes the authentication challenge is a human-operated client.

Techniques for optimizing authentication challenges for detection of malicious attacks

Breaking text-based CAPTCHAs with variable word and character orientation

Websites present users with puzzles called CAPTCHAs to curb abuse caused by computer algorithms masquerading as people. While CAPTCHAs are generally effective at stopping abuse, they might impair website usability if they are not properly designed. In this paper we describe how we designed two new CAPTCHA schemes for Google that focus on maximizing usability. We began by running an evaluation on Amazon Mechanical Turk with over 27,000 respondents to test the usability of different feature combinations. Then we studied user preferences using Google's consumer survey infrastructure. Finally, drawing on the insights gleaned during those studies, we tested our new captcha schemes first on Mechanical Turk and then on a fraction of production traffic. The resulting scheme is now an integral part of our production system and is served to millions of users. Our scheme achieved a 95.3% human accuracy, a 6.7.

/pdf/easy-does-it-more-usable-captchas-4atxdz85dt.pdf

Easy does it: more usable CAPTCHAs

Systems and methods for verifying a user based on reputational information are provided. In particular, a computerized CAPTCHA system consisting of one or more computers can determine a trust score based on one or more reputation signals associated with a user computing device, select a challenge to provide to the user computing device based on the trust score, and determine whether to verify the user computing device based on a received response to the challenge and/or the trust score.

Systems and methods for verifying a user based on reputational information

Computerized CAPTCHA systems using a direct connection with user computing devices are provided. An example computerized CAPTCHA system is configured to perform operations. The operations include receiving a request from a user computing device to engage in a verification process. The request is received independent of a resource provider from which the user computing device has requested a resource. The operations include providing a challenge to the user computing device at least in part in response to the request for engagement in the verification process and receiving a response to the challenge from the user computing device. The operations include determining whether the user computing device should be verified based at least in part on the response and providing a verification token to the user computing device when it is determined that the user computing device should be verified.

Angelique Moscicki

Papers

Easy does it: more usable CAPTCHAs

Systems and methods for verifying a user based on reputational information

Computerized CAPTCHA Systems Using A Direct Connection With User Computing Device