Capture the Bot: Using Adversarial Examples to Improve CAPTCHA Robustness to Bot Attacks

The researchers propose CAPTURE, a new CAPTCHA scheme based on adversarial examples, which repurposes adversarial examples, typically used to mislead machine learning models, for defense. Experimental evaluation shows that CAPTCHAs generated by CAPTURE are easy for humans to solve while effectively blocking machine learning-based bot cracking programs.