I found an interesting study "Breaking reCAPTCHAv2", which examines the effectiveness of using advanced machine learning methods to solve captchas: ? reCAPTCHAv2 from Google.
Using the YOLOv8 model, it was possible (https://t.me/russian_osint) to solve 100% of CAPTCHA problems based on reCAPTCHAv2. In previous cases, the result was much more modest - only 68-71%.
Advanced bots and humans (https://t.me/russian_osint) try approximately the same number of attempts to solve CAPTCHAs, and in some cases bots even outperform humans.
According to the authors, the main conclusion of the study is that CAPTCHA systems such as reCAPTCHAv2 can no longer reliably distinguish humans from machines. Methods of protection against bots need to be reconsidered.
? https://arxiv.org/pdf/2409.08831