If you’re tired of the days of completing captcha tests to prove you’re not a robot, you’re not alone. Now, it looks like reCAPTCHAv2, which is probably best known as the latest version to directly test image recognition, can be defeated by current generation AI models with a 100% success rate. Using an existing You Only Look Once (YOLO) object recognition model after training it on 14,000 labeled traffic images, the research paper titled “Breaking reCAPTCHAv2” was submitted to arXiv on September 13. I was able to defeat reCAPTCHAv2 with 100% success rate. .
So what does this mean for today’s internet users and website operators? It depends! As it turns out, Google’s reCAPTCHAv2 is actually a bit outdated compared to reCAPTCHAv3, which uses other metrics to determine whether a user is human, rather than directly testing the user with an image recognition challenge. …unless your web host chooses to enable this feature. There are potential false positives with reCAPTCHAv3, which in theory should be mitigated by the ability to fall back to reCAPTCHAv2 tests…but it’s now common knowledge that reCAPTCHAv2 is disabled. Things can change faster than we expect.
As the original paper concludes, “By conducting systematic experiments, we show that automated systems using advanced AI technologies, such as the YOLO model, can successfully solve image-based capture. (…) This finding shows that the reliability of image-based capture as a definitive method for distinguishing between humans and bots is not affected by the rapidly advancing field of artificial intelligence, as current capture mechanisms It shows.”
After discussing how future research can be improved, the paper goes on to say that “Google’s use of reCAPTCHAv2 improves the security of websites on the Internet by better distinguishing between real users and automated bots.” For example, reCAPTCHAV2 has played an important role in improving advertising revenue diversion and competitive Addresses the problem of scraping (…) by preventing automated theft aimed at gaining an advantage. , and the large amounts of data needed to train them. ”
In other words, this research was not done purely to compensate for the inadequacies of reCAPTCHAv2 in the face of AI’s incredible capabilities. If anything, the researchers conclude, the existence of strong, functioning captcha systems and similar systems is a good thing, if not “essential” to a healthy Internet future. And they are right. While the paper’s introduction asserts that “we have now officially moved beyond capture,” the conclusion argues that “capture technology needs to evolve aggressively to stay ahead of the rapid enhancements of AI.” ” has been confirmed.
