University of Maryland researchers warn that with limited resources, threat actors could launch a successful cyberattack on Google’s bot-detecting reCaptcha service.
In an academic paper detailing their findings, the researchers discuss how they created a tool called unCaptcha, which uses audio files in conjunction with artificial intelligence (AI) technologies such as speech-to-text software to bypass the Google security mechanism.
Over more than 450 tests, the unCaptcha tool defeated reCaptcha with 85 percent accuracy in 5.42 seconds, on average. This study proved that threat actors could potentially break into web-based services, pursue automated account creation and more.
How Researchers Got Around reCaptcha
Online users will recognize reCaptcha as a small box that appears on many websites when signing up or logging in to digital services. Website visitors are typically asked to solve a challenge to prove they’re human, whether it’s typing in letters next to a distorted rendering of the letters, answering a question or clicking on images.
In this case, the University of Maryland researchers took advantage of the fact that Google’s system offers an audio version of its challenges for those who may be visually impaired. The attack method involved navigating to Google’s reCaptcha demo site, finding the audio challenge and downloading it, then putting it through a speech-to-text engine. After an answer had been parsed, it could be typed in and submitted.
While Google initially responded by creating a new version of reCaptcha, the researchers did the same thing with unCaptcha and were even more successful. In an interview with BleepingComputer, one of the researchers said the new version had a success rate of around 91 percent after more than 600 attempts.
Securing the Web Without CAPTCHAs
The research paper recommends a number of possible countermeasures to a tool such as unCaptcha, including broadening the sound bytes of reCaptcha audio challenges and adding distortion. CAPTCHAs are far from the only option available to protect digital services, however.
IBM Security experts, for example, discussed the promise of managed identity and access management (IAM), which allows organizations to not only protect online services with additional layers of security, but also have a third party deal with operational chores such as patching and resolving upcoming incidents. If a group of academics can automate attacks on CAPTCHA systems this successfully, it may be time for security leaders and their teams to look for something more sophisticated.