Researchers at ETH Zurich trained an AI model that can effectively solve CAPTCHA puzzles, indicating a significant challenge for the usefulness of CAPTCHA in distinguishing human users from bots.
The YOLO model's adaptations allowed it to crack Google's reCAPTCHA v2 with a success rate of 100%, outpacing previous attempts that only achieved 68 to 71 percent.
The study highlighted that mimicking human behavior is essential for the AI to evade detection; the model takes similar attempts as humans, complicating the CAPTCHA system's effectiveness.
One core finding suggested that reCAPTCHA v2's reliance on tools like browser history and cookies might be exploited by bots imitating human web interactions.
Collection
[
|
...
]