AI Beats 100% of Traffic-Image CAPTCHAs
New research from ETH Zurich shows that bots using advanced image-recognition models can now achieve a 100% success rate in solving Google’s reCAPTCHA v2, a common security measure that challenges users to identify objects in images to prove they are human. Despite Google’s shift to the more sophisticated reCAPTCHA v3, millions of websites still rely on the older v2 system. The research team fine-tuned the YOLO object-recognition model to accurately solve these CAPTCHA challenges, using additional tactics like VPNs and simulated human-like mouse movements to avoid detection. Their bot could solve CAPTCHA images as accurately as humans, and even slightly faster in some cases.
This breakthrough highlights the declining effectiveness of traditional CAPTCHAs as AI advances. The study underscores how bots can now mimic human behaviors in tasks once thought to be uniquely human, pushing security measures towards more subtle user identification methods, such as device fingerprinting. Google acknowledged this shift, noting that reCAPTCHA v3 is already in use on millions of websites, offering invisible protection without visual challenges. However, as AI continues to improve, distinguishing humans from bots may become increasingly difficult.