AI bots are now smart enough to solve image-based CAPTCHAs with a 100-percent success rate.
New research led by Andreas Plesner, a PhD student at ETH Zurich, revealed that bots running locally-run artificial intelligence are capable of human-level accuracy in solving Google’s ReCAPTCHA v2, a system designed to distinguish human users from automated programs to safeguard websites from abuse.
Using the YOLO (You Only Look Once) object-recognition model with 14,000 labelled traffic images as training data, the researchers developed a bot that was able to identify objects in CAPTCHA images with 69 to 100 percent accuracy, depending on the objects.
The latest version of Google’s reCAPTCHA is v3, which is designed to be invisible and allow users to interact with websites automatically. However, ReCAPTCHA v2 is still common and typically used as a fallback option to disrupt users when it detects abnormal user behavior, prompting them to solve an image identification challenge. With this in mind, the researchers made use of VPNs, simulated mouse movements, and other techniques to enhance the effectiveness of their bot in beating ReCAPTCHA.
Before their findings, the researchers said previous academic studies on solving CAPTCHAs only had an accuracy between 68 and 71 percent. With the rising popularity of large language models and the automated data scraping that entails to train these models, the researchers emphasized the need for CAPTCHA technologies to evolve, to outsmart AI, and keep online environments secure.
You can read the research paper in full at arXiv.org.
This article, AI bots can now solve image CAPTCHAs with 100% success rate, was originally published at NoypiGeeks | Philippines Technology News, Reviews and How to's.
Comments
Post a Comment