The unCaptcha automated system can once again bypass Google’s reCAPTCHA challenges, despite major updates to the security service.
The unCaptcha system was created in April 2017 to target Google’s reCAPTCHA and similar security systems that aim at protecting websites from spam and abuse, and was able to defeat those with high accuracy.
Soon after the system was revealed by computer science experts from the University of Maryland (UM), Google updated its security service to mitigate the attack, but unCaptcha has been updated to bypass the mitigations.
The unCaptcha system was designed to target the audio challenges presented by reCAPTCHA, but the security experts who designed it revealed that it could defeat other systems as well, including BotDetect, Yahoo, and PayPal image challenges, all with high accuracy.
To mitigate the attack, Google improved reCAPTCHA’s browser automation detection, and also switched from spoken digits to spoken phrases, two major changes that were successful in protecting against the original unCaptcha attack.
In June 2018, however, an updated unCaptcha was able to solve the new challenges, effectively bypassing Google’s security service once again. What’s more, the system’s accuracy and effectiveness have also improved.
“Thanks to the changes to the audio challenge, passing ReCaptcha is easier than ever before. The code now only needs to make a single request to a free, publicly available speech to text API to achieve around 90% accuracy over all captchas,” the security experts behind the system explain.
reCAPTCHA, however, can no longer be targeted using the Selenium browser automation engine, and unCaptcha switched to a screen clicker to move to certain pixels on the screen to mimic human movement on a page.
“There is certainly work to be done here – the coordinates need to be updated for each new user and is not the most robust,” the researchers say.
The updated system works in a pretty straightforward manner: it navigates to Google’s ReCaptcha Demo site and then to the audio challenge, downloads the challenge, submits it to Speech To Text, parses the response and types the answer, and clicks submit and checks if the attack was successful.
The reCAPTCHA team was informed on the discovery in June 2018 and a formal issue was opened on June 27, 2018. The experts decided to wait for six months before making their findings public, to give Google enough time to address the underlying architectural issues.
“The Recaptcha team is aware of this attack vector, and have confirmed they are okay with us releasing this code, despite its current success rate. This attack vector was deemed out of scope for the bug bounty program,” the experts also note.
In October 2018, Google announced the launch of reCAPTCHA v3, with a focus on improving user experience by running adaptive risk analysis in the background and providing a score to inform website owners of how suspicious an interaction is.
Meant to be a proof-of-concept, the new unCaptcha variant won’t be updated when Google makes changes to its service. This means that the code is likely to break at any time. The experts also removed their API keys from all the necessary queries.
Related: Automated System Defeats reCAPTCHA With High Accuracy
Related: Google Launches reCAPTCHA v3