14 DAY TRIAL // British and Chinese researchers from UK's Lancaster University and China's Northwest University and Peking University have developed a new artificial intelligence-based attack against text captchas which could make them obsolete.
"The new algorithm, based on deep learning methods, is the most effective solver of captcha security and authentication systems to date and is able to defeat versions of text captcha schemes used to defend the majority of the world’s most popular websites," states the research team.
According to the research team behind the "Yet Another Text Captcha Solver: A Generative Adversarial Network Based Approach" paper, their new captcha solver does not need high volumes of real captchas as a training ground since it is not based on a machine-learning approach.
"This is achieved by first learning a captcha synthesizer to automatically generate synthetic captchas to learn a base solver, and then fine-tuning the base solver on a small set of real captchas using transfer learning," says the research paper's abstract.
Moreover, the use of artificial intelligence makes the new captcha solver a lot less time consuming and labor intensive which makes it a viable solution even for consumer-grade computing systems.
The new algorithm can solve captchas in under 0.05 seconds using a GPU
The solver algorithm uses a Generative Adversarial Network (GAN) technique which generates vast numbers of training captchas almost identical to the authentic ones with the help of a generator program.
In turn, the generated captchas will be used to train the solver algorithm, later tested and refined against real captchas from popular websites such as Microsoft, Wikipedia, eBay, and Google to obtain notably higher accuracy when compared to other captcha attack systems.
The research team showed in the study that their new captcha solver is very efficient considering that it can be used to solve captchas in up to 0.05 seconds with the help of a desktop GPU.
Furthermore, they were also able to show that the newly devised captcha attack can be used to bypass website security systems utilizing most modern text captcha schemes.
"Given that deep learning and generative adversarial based approaches are making great progress in solving image-related tasks, the insights provided in this work can help security experts to revisit the design and usability of text captchas," concludes the study.
