An attack method discovered in 2017 for defeating the audio version of Google’s reCAPTCHA system using speech-to-text services has once again been resurrected.
A team of researchers from the University of Maryland showed in 2017 that online speech-to-text services could be used to automatically solve reCAPTCHA v2 audio challenges with a high degree of accuracy. The attack was named by the researchers unCaptcha.
After the method was disclosed, Google made some changes to its reCAPTCHA system and unCaptcha no longer worked. However, in January 2019, researchers announced that they had managed to revive the attack — they dubbed it unCaptcha2 — and they released proof-of-concept (PoC) code with Google’s permission.
At the time, the researchers noted that they would not be updating their code and that it would likely stop working at some point.
As expected, the PoC did stop working, but Germany-based researcher Nikolai Tschacher has managed to tweak the PoC for unCaptcha2 to make it work against the latest version of reCAPTCHA v2. Tschacher has published a video showing how a bot can solve the audio reCAPTCHA using Google’s own speech-to-text API with an accuracy of 97%.
Google introduced reCAPTCHA v3 in 2018, which improves user experience by running adaptive risk analysis in the background rather than displaying challenges, but Tschacher pointed out that “reCAPTCHA v2 is still used in the new reCAPTCHA v3 as a fall-back mechanism.”
The researcher has published PoC code, along with an explanation of the changes he made in unCaptcha3 compared to unCaptcha2.
It’s worth noting that others have created free web browser extensions that help users automatically solve reCAPTCHA challenges with the press of a button using the unCaptcha method.
SecurityWeek has reached out to Google for comment and will update this article if the company responds.
Related: Researcher Discloses Google ReCaptcha v2 Bypass
Related: Researchers Devise Scalable Attack Against Google, Facebook reCaptcha