Since the first CAPTCHA was created more than 20 years ago, new methods for telling computers and humans started appearing. From the basic text-based CAPTCHAs, which required only writing letters visible on the picture to advanced game-based CAPTCHAs which require user to play short game or no CAPTCHA reCAPTCHA which analyses actions of the users to determine if they are humans. In this work there is described a new possible type of CAPTCHA, Composed Image-Text CAPTCHA, connecting the image and text in one task to make it more robust against attacks. The proposed CAPTCHA looks like an image CAPTCHA, but every of the images additionally contains a text, which may be related to the given image. The task for the user is to choose the images mentioned in the question, but only the ones in which the text matches the image as well. The work contains a description of how it should work and how AI could be used to generate it. Possible advantages and disadvantages will be discussed.