Artificial intelligence was able to pass the "Fully automated public Turing test to distinguish a computer from a human" or simply CAPTCHA. This was discovered by scientists at the Swiss Federal Institute of Technology Zurich, TechRadar reports.
Researchers have developed their own bot based on a widely used AI model for image processing called You Only Look Once (YOLO). The scientists adapted YOLO to the popular reCAPTCHAv2 version of Google's CAPTCHAv2. This is the same "captcha" that asks the user to click on an image of a car, bicycle, bridge, or traffic light.
The researchers ran 14,000 photos of city streets through the model, and this was enough to pass most of the tests. AI did not solve all puzzles perfectly the first time. When it made a mistake, YOLO remembered it and corrected itself. According to the research team, the simplicity of the system played in favor of AI. The tests do not have a large variety of objects to identify.
The fact that AI can now bypass CAPTCHAs is a wake-up call for cybersecurity. CAPTCHA was a simple but reliable defense against bots engaging in activities such as spamming, creating fake accounts, or DDoS attacks.