The end of captchas? Artificial intelligence surpasses them 100%

The artificial intelligence (AI) has managed to overcome one of the most common security barriers on the Internet: captchas. A study recent has shown that current language models can solve 100% of the challenges posed by Google’s reCAPTCHA v2, thus questioning the effectiveness of this method to distinguish between humans and bots.

The report, published on the arXiv platform, notes that the researchers used an image recognition model known as YOLOtrained with 14,000 traffic images. With this model, the researchers managed to overcome all the challenges of reCAPTCHA v2, making it clear that traditional security techniques are no longer an obstacle to advanced AI technology.

Although this version is a little outdated compared to v3, which uses other types of metrics to determine if we are human, it is certainly paradoxical to know that this test can be completed by a machine.

Captchas have been a resource for years to combat malicious activities such as “scraping” or automated data theft. Its implementation has been crucial to protect advertising revenue and prevent data exploitation by bots. However, the study highlights some versions of captchas that are currently used, although effective in the past, have become obsolete in the face of advances in artificial intelligence.

The study also addresses how AI has challenged the effectiveness of image-based captchas. “We have shown that automated systems using advanced AI technologies, such as YOLO models, can successfully solve image-based captchas,” the researchers note. This reality raises doubts about the reliability of these methods as a definitive mechanism to distinguish between humans and machines.

Despite the achievements of AI, the authors acknowledge that captchas have played an important role in internet security. However, they also warn that We are entering “the era beyond captchas”. To stay relevant, security methods must proactively evolve to stay ahead of rapid improvements in AI.

The study ends by raising the need to develop new verification techniques. “Captchas must evolve if they are to remain an effective method of online security,” the researchers conclude. Meanwhile, the race between security measures and AI continues, leaving the question in the air: What will be the next solution to keep the internet safe?